mrq/stable-diffusion-webui
1
1
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'upstream/master' into PowerShell

Browse Source
This commit is contained in:
d8ahazard 2023-01-10 21:31:50 -06:00
commit e0eabd9db1
75 changed files with 3139 additions and 1303 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.github/ISSUE_TEMPLATE/feature_request.yml vendored
View File

@ -1,7 +1,7 @@
name: Feature request
description: Suggest an idea for this project
title: "[Feature Request]: "
labels: ["suggestion"]
labels: ["enhancement"]
body:
- type: checkboxes

4
.github/PULL_REQUEST_TEMPLATE/pull_request_template.md → .github/pull_request_template.md vendored
View File

@ -18,8 +18,8 @@ More technical discussion about your changes go here, plus anything that a maint
List the environment you have developed / tested this on. As per the contributing page, changes should be able to work on Windows out of the box.
- OS: [e.g. Windows, Linux]
- Browser [e.g. chrome, safari]
- Graphics card [e.g. NVIDIA RTX 2080 8GB, AMD RX 6600 8GB]
- Browser: [e.g. chrome, safari]
- Graphics card: [e.g. NVIDIA RTX 2080 8GB, AMD RX 6600 8GB]
**Screenshots or videos of your changes**

13
.github/workflows/on_pull_request.yaml vendored
View File

@ -19,22 +19,19 @@ jobs:
- name: Checkout Code
uses: actions/checkout@v3
- name: Set up Python 3.10
uses: actions/setup-python@v3
uses: actions/setup-python@v4
with:
python-version: 3.10.6
- uses: actions/cache@v2
with:
path: ~/.cache/pip
key: ${{ runner.os }}-pip-${{ hashFiles('**/requirements.txt') }}
restore-keys: |
${{ runner.os }}-pip-
cache: pip
cache-dependency-path: |
**/requirements*txt
- name: Install PyLint
run: |
python -m pip install --upgrade pip
pip install pylint
# This lets PyLint check to see if it can resolve imports
- name: Install dependencies
run : |
run: |
export COMMANDLINE_ARGS="--skip-torch-cuda-test --exit"
python launch.py
- name: Analysing the code with pylint

10
.github/workflows/run_tests.yaml vendored
View File

@ -14,13 +14,11 @@ jobs:
uses: actions/setup-python@v4
with:
python-version: 3.10.6
- uses: actions/cache@v3
with:
path: ~/.cache/pip
key: ${{ runner.os }}-pip-${{ hashFiles('**/requirements.txt') }}
restore-keys: ${{ runner.os }}-pip-
cache: pip
cache-dependency-path: |
**/requirements*txt
- name: Run tests
run: python launch.py --tests basic_features --no-half --disable-opt-split-attention --use-cpu all --skip-torch-cuda-test
run: python launch.py --tests --no-half --disable-opt-split-attention --use-cpu all --skip-torch-cuda-test
- name: Upload main app stdout-stderr
uses: actions/upload-artifact@v3
if: always()

12
README.md
View File

@ -1,9 +1,7 @@
# Stable Diffusion web UI
A browser interface based on Gradio library for Stable Diffusion.
![](txt2img_Screenshot.png)
Check the [custom scripts](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Custom-Scripts) wiki page for extra scripts developed by users.
![](screenshot.png)
## Features
[Detailed feature showcase with images](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features):
@ -97,9 +95,8 @@ Alternatively, use online services (like Google Colab):
1. Install [Python 3.10.6](https://www.python.org/downloads/windows/), checking "Add Python to PATH"
2. Install [git](https://git-scm.com/download/win).
3. Download the stable-diffusion-webui repository, for example by running `git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git`.
4. Place `model.ckpt` in the `models` directory (see [dependencies](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Dependencies) for where to get it).
5. _*(Optional)*_ Place `GFPGANv1.4.pth` in the base directory, alongside `webui.py` (see [dependencies](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Dependencies) for where to get it).
6. Run `webui-user.bat` from Windows Explorer as normal, non-administrator, user.
4. Place stable diffusion checkpoint (`model.ckpt`) in the `models/Stable-diffusion` directory (see [dependencies](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Dependencies) for where to get it).
5. Run `webui-user.bat` from Windows Explorer as normal, non-administrator, user.
### Automatic Installation on Linux
1. Install the dependencies:
@ -127,6 +124,8 @@ Here's how to add code to this repo: [Contributing](https://github.com/AUTOMATIC
The documentation was moved from this README over to the project's [wiki](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki).
## Credits
Licenses for borrowed code can be found in `Settings -> Licenses` screen, and also in `html/licenses.html` file.
- Stable Diffusion - https://github.com/CompVis/stable-diffusion, https://github.com/CompVis/taming-transformers
- k-diffusion - https://github.com/crowsonkb/k-diffusion.git
- GFPGAN - https://github.com/TencentARC/GFPGAN.git
@ -139,6 +138,7 @@ The documentation was moved from this README over to the project's [wiki](https:
- Ideas for optimizations - https://github.com/basujindal/stable-diffusion
- Cross Attention layer optimization - Doggettx - https://github.com/Doggettx/stable-diffusion, original idea for prompt editing.
- Cross Attention layer optimization - InvokeAI, lstein - https://github.com/invoke-ai/InvokeAI (originally http://github.com/lstein/stable-diffusion)
- Sub-quadratic Cross Attention layer optimization - Alex Birch (https://github.com/Birch-san/diffusers/pull/1), Amin Rezaei (https://github.com/AminRezaei0x443/memory-efficient-attention)
- Textual Inversion - Rinon Gal - https://github.com/rinongal/textual_inversion (we're not using his code, but we are using his ideas).
- Idea for SD upscale - https://github.com/jquesnelle/txt2imghd
- Noise generation for outpainting mk2 - https://github.com/parlance-zz/g-diffuser-bot

13
html/footer.html Normal file
View File

@ -0,0 +1,13 @@
<div>
<a href="/docs">API</a>
 • 
<a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui">Github</a>
 • 
<a href="https://gradio.app">Gradio</a>
 • 
<a href="/" onclick="javascript:gradioApp().getElementById('settings_restart_gradio').click(); return false">Reload UI</a>
</div>
<br />
<div class="versions">
{versions}
</div>

419
html/licenses.html Normal file
View File

@ -0,0 +1,419 @@
<style>
#licenses h2 {font-size: 1.2em; font-weight: bold; margin-bottom: 0.2em;}
#licenses small {font-size: 0.95em; opacity: 0.85;}
#licenses pre { margin: 1em 0 2em 0;}
</style>
<h2><a href="https://github.com/sczhou/CodeFormer/blob/master/LICENSE">CodeFormer</a></h2>
<small>Parts of CodeFormer code had to be copied to be compatible with GFPGAN.</small>
<pre>
S-Lab License 1.0
Copyright 2022 S-Lab
Redistribution and use for non-commercial purpose in source and
binary forms, with or without modification, are permitted provided
that the following conditions are met:
1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in
the documentation and/or other materials provided with the
distribution.
3. Neither the name of the copyright holder nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
In the event that redistribution and/or use for commercial purpose in
source or binary forms, with or without modification is required,
please contact the contributor(s) of the work.
</pre>
<h2><a href="https://github.com/victorca25/iNNfer/blob/main/LICENSE">ESRGAN</a></h2>
<small>Code for architecture and reading models copied.</small>
<pre>
MIT License
Copyright (c) 2021 victorca25
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
</pre>
<h2><a href="https://github.com/xinntao/Real-ESRGAN/blob/master/LICENSE">Real-ESRGAN</a></h2>
<small>Some code is copied to support ESRGAN models.</small>
<pre>
BSD 3-Clause License
Copyright (c) 2021, Xintao Wang
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
1. Redistributions of source code must retain the above copyright notice, this
list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice,
this list of conditions and the following disclaimer in the documentation
and/or other materials provided with the distribution.
3. Neither the name of the copyright holder nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
</pre>
<h2><a href="https://github.com/invoke-ai/InvokeAI/blob/main/LICENSE">InvokeAI</a></h2>
<small>Some code for compatibility with OSX is taken from lstein's repository.</small>
<pre>
MIT License
Copyright (c) 2022 InvokeAI Team
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
</pre>
<h2><a href="https://github.com/Hafiidz/latent-diffusion/blob/main/LICENSE">LDSR</a></h2>
<small>Code added by contirubtors, most likely copied from this repository.</small>
<pre>
MIT License
Copyright (c) 2022 Machine Vision and Learning Group, LMU Munich
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
</pre>
<h2><a href="https://github.com/pharmapsychotic/clip-interrogator/blob/main/LICENSE">CLIP Interrogator</a></h2>
<small>Some small amounts of code borrowed and reworked.</small>
<pre>
MIT License
Copyright (c) 2022 pharmapsychotic
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
</pre>
<h2><a href="https://github.com/JingyunLiang/SwinIR/blob/main/LICENSE">SwinIR</a></h2>
<small>Code added by contributors, most likely copied from this repository.</small>
<pre>
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [2021] [SwinIR Authors]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
</pre>
<h2><a href="https://github.com/AminRezaei0x443/memory-efficient-attention/blob/main/LICENSE">Memory Efficient Attention</a></h2>
<small>The sub-quadratic cross attention optimization uses modified code from the Memory Efficient Attention package that Alex Birch optimized for 3D tensors. This license is updated to reflect that.</small>
<pre>
MIT License
Copyright (c) 2023 Alex Birch
Copyright (c) 2023 Amin Rezaei
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
</pre>

10
javascript/dragdrop.js vendored
View File

@ -9,11 +9,19 @@ function dropReplaceImage( imgWrap, files ) {
return;
}
const tmpFile = files[0];
imgWrap.querySelector('.modify-upload button + button, .touch-none + div button + button')?.click();
const callback = () => {
const fileInput = imgWrap.querySelector('input[type="file"]');
if ( fileInput ) {
fileInput.files = files;
if ( files.length === 0 ) {
files = new DataTransfer();
files.items.add(tmpFile);
fileInput.files = files.files;
} else {
fileInput.files = files;
}
fileInput.dispatchEvent(new Event('change'));
}
};

19
javascript/hints.js
View File

@ -4,7 +4,7 @@ titles = {
"Sampling steps": "How many times to improve the generated image iteratively; higher values take longer; very low values can produce bad results",
"Sampling method": "Which algorithm to use to produce the image",
"GFPGAN": "Restore low quality faces using GFPGAN neural network",
"Euler a": "Euler Ancestral - very creative, each can get a completely different picture depending on step count, setting steps to higher than 30-40 does not help",
"Euler a": "Euler Ancestral - very creative, each can get a completely different picture depending on step count, setting steps higher than 30-40 does not help",
"DDIM": "Denoising Diffusion Implicit Models - best at inpainting",
"DPM adaptive": "Ignores step count - uses a number of steps determined by the CFG and resolution",
@ -74,16 +74,13 @@ titles = {
"Style 1": "Style to apply; styles have components for both positive and negative prompts and apply to both",
"Style 2": "Style to apply; styles have components for both positive and negative prompts and apply to both",
"Apply style": "Insert selected styles into prompt fields",
"Create style": "Save current prompts as a style. If you add the token {prompt} to the text, the style use that as placeholder for your prompt when you use the style in the future.",
"Create style": "Save current prompts as a style. If you add the token {prompt} to the text, the style uses that as a placeholder for your prompt when you use the style in the future.",
"Checkpoint name": "Loads weights from checkpoint before making images. You can either use hash or a part of filename (as seen in settings) for checkpoint name. Recommended to use with Y axis for less switching.",
"Inpainting conditioning mask strength": "Only applies to inpainting models. Determines how strongly to mask off the original image for inpainting and img2img. 1.0 means fully masked, which is the default behaviour. 0.0 means a fully unmasked conditioning. Lower values will help preserve the overall composition of the image, but will struggle with large changes.",
"vram": "Torch active: Peak amount of VRAM used by Torch during generation, excluding cached data.\nTorch reserved: Peak amount of VRAM allocated by Torch, including all active and cached data.\nSys VRAM: Peak amount of VRAM allocation across all applications / total GPU VRAM (peak utilization%).",
"Highres. fix": "Use a two step process to partially create an image at smaller resolution, upscale, and then improve details in it without changing composition",
"Scale latent": "Uscale the image in latent space. Alternative is to produce the full image from latent representation, upscale that, and then move it back to latent space.",
"Eta noise seed delta": "If this values is non-zero, it will be added to seed and used to initialize RNG for noises when using samplers with Eta. You can use this to produce even more variation of images, or you can use this to match images of other software if you know what you are doing.",
"Do not add watermark to images": "If this option is enabled, watermark will not be added to created images. Warning: if you do not add watermark, you may be behaving in an unethical manner.",
@ -95,12 +92,18 @@ titles = {
"Weighted sum": "Result = A * (1 - M) + B * M",
"Add difference": "Result = A + (B - C) * M",
"Learning rate": "how fast should the training go. Low values will take longer to train, high values may fail to converge (not generate accurate results) and/or may break the embedding (This has happened if you see Loss: nan in the training info textbox. If this happens, you need to manually restore your embedding from an older not-broken backup).\n\nYou can set a single numeric value, or multiple learning rates using the syntax:\n\n rate_1:max_steps_1, rate_2:max_steps_2, ...\n\nEG: 0.005:100, 1e-3:1000, 1e-5\n\nWill train with rate of 0.005 for first 100 steps, then 1e-3 until 1000 steps, then 1e-5 for all remaining steps.",
"Learning rate": "How fast should training go. Low values will take longer to train, high values may fail to converge (not generate accurate results) and/or may break the embedding (This has happened if you see Loss: nan in the training info textbox. If this happens, you need to manually restore your embedding from an older not-broken backup).\n\nYou can set a single numeric value, or multiple learning rates using the syntax:\n\n rate_1:max_steps_1, rate_2:max_steps_2, ...\n\nEG: 0.005:100, 1e-3:1000, 1e-5\n\nWill train with rate of 0.005 for first 100 steps, then 1e-3 until 1000 steps, then 1e-5 for all remaining steps.",
"Clip skip": "Early stopping parameter for CLIP model; 1 is stop at last layer as usual, 2 is stop at penultimate layer, etc.",
"Approx NN": "Cheap neural network approximation. Very fast compared to VAE, but produces pictures with 4 times smaller horizontal/vertical resoluton and lower quality.",
"Approx cheap": "Very cheap approximation. Very fast compared to VAE, but produces pictures with 8 times smaller horizontal/vertical resoluton and extremely low quality."
"Approx NN": "Cheap neural network approximation. Very fast compared to VAE, but produces pictures with 4 times smaller horizontal/vertical resolution and lower quality.",
"Approx cheap": "Very cheap approximation. Very fast compared to VAE, but produces pictures with 8 times smaller horizontal/vertical resolution and extremely low quality.",
"Hires. fix": "Use a two step process to partially create an image at smaller resolution, upscale, and then improve details in it without changing composition",
"Hires steps": "Number of sampling steps for upscaled picture. If 0, uses same as for original.",
"Upscale by": "Adjusts the size of the image by multiplying the original width and height by the selected value. Ignored if either Resize width to or Resize height to are non-zero.",
"Resize width to": "Resizes image to this width. If 0, width is inferred from either of two nearby sliders.",
"Resize height to": "Resizes image to this height. If 0, height is inferred from either of two nearby sliders."
}

25
javascript/hires_fix.js Normal file
View File

@ -0,0 +1,25 @@
function setInactive(elem, inactive){
console.log(elem)
if(inactive){
elem.classList.add('inactive')
} else{
elem.classList.remove('inactive')
}
}
function onCalcResolutionHires(enable, width, height, hr_scale, hr_resize_x, hr_resize_y){
console.log(enable, width, height, hr_scale, hr_resize_x, hr_resize_y)
hrUpscaleBy = gradioApp().getElementById('txt2img_hr_scale')
hrResizeX = gradioApp().getElementById('txt2img_hr_resize_x')
hrResizeY = gradioApp().getElementById('txt2img_hr_resize_y')
gradioApp().getElementById('txt2img_hires_fix_row2').style.display = opts.use_old_hires_fix_width_height ? "none" : ""
setInactive(hrUpscaleBy, opts.use_old_hires_fix_width_height || hr_resize_x > 0 || hr_resize_y > 0)
setInactive(hrResizeX, opts.use_old_hires_fix_width_height || hr_resize_x == 0)
setInactive(hrResizeY, opts.use_old_hires_fix_width_height || hr_resize_y == 0)
return [enable, width, height, hr_scale, hr_resize_x, hr_resize_y]
}

4
javascript/imageviewer.js
View File

@ -148,8 +148,8 @@ function showGalleryImage() {
if(e && e.parentElement.tagName == 'DIV'){
e.style.cursor='pointer'
e.style.userSelect='none'
e.addEventListener('click', function (evt) {
if(!opts.js_modal_lightbox) return;
e.addEventListener('mousedown', function (evt) {
if(!opts.js_modal_lightbox || evt.button != 0) return;
modalZoomSet(gradioApp().getElementById('modalImage'), opts.js_modal_lightbox_initially_zoomed)
showModal(evt)
}, true);

17
javascript/ui.js
View File

@ -1,4 +1,4 @@
// various functions for interation with ui.py not large enough to warrant putting them in separate files
// various functions for interaction with ui.py not large enough to warrant putting them in separate files
function set_theme(theme){
gradioURL = window.location.href
@ -19,7 +19,7 @@ function selected_gallery_index(){
function extract_image_from_gallery(gallery){
if(gallery.length == 1){
return gallery[0]
return [gallery[0]]
}
index = selected_gallery_index()
@ -28,7 +28,7 @@ function extract_image_from_gallery(gallery){
return [null]
}
return gallery[index];
return [gallery[index]];
}
function args_to_array(args){
@ -188,6 +188,17 @@ onUiUpdate(function(){
img2img_textarea = gradioApp().querySelector("#img2img_prompt > label > textarea");
img2img_textarea?.addEventListener("input", () => update_token_counter("img2img_token_button"));
}
show_all_pages = gradioApp().getElementById('settings_show_all_pages')
settings_tabs = gradioApp().querySelector('#settings div')
if(show_all_pages && settings_tabs){
settings_tabs.appendChild(show_all_pages)
show_all_pages.onclick = function(){
gradioApp().querySelectorAll('#settings > div').forEach(function(elem){
elem.style.display = "block";
})
}
}
})
let txt2img_textarea, img2img_textarea = undefined;

21
launch.py
View File

@ -13,6 +13,21 @@ dir_extensions = "extensions"
python = sys.executable
git = os.environ.get('GIT', "git")
index_url = os.environ.get('INDEX_URL', "")
stored_commit_hash = None
def commit_hash():
global stored_commit_hash
if stored_commit_hash is not None:
return stored_commit_hash
try:
stored_commit_hash = run(f"{git} rev-parse HEAD").strip()
except Exception:
stored_commit_hash = "<none>"
return stored_commit_hash
def extract_arg(args, name):
@ -194,10 +209,7 @@ def prepare_environment():
xformers = '--xformers' in sys.argv
ngrok = '--ngrok' in sys.argv
try:
commit = run(f"{git} rev-parse HEAD").strip()
except Exception:
commit = "<none>"
commit = commit_hash()
print(f"Python {sys.version}")
print(f"Commit hash: {commit}")
@ -270,6 +282,7 @@ def tests(test_dir):
print(f"Launching Web UI in another process for testing with arguments: {' '.join(sys.argv[1:])}")
os.environ['COMMANDLINE_ARGS'] = ""
with open('test/stdout.txt', "w", encoding="utf8") as stdout, open('test/stderr.txt', "w", encoding="utf8") as stderr:
proc = subprocess.Popen([sys.executable, *sys.argv], stdout=stdout, stderr=stderr)

147
modules/api/api.py
View File

@ -1,24 +1,25 @@
import base64
import io
import time
import datetime
import uvicorn
from threading import Lock
from io import BytesIO
from gradio.processing_utils import decode_base64_to_file
from fastapi import APIRouter, Depends, FastAPI, HTTPException
from fastapi import APIRouter, Depends, FastAPI, HTTPException, Request, Response
from fastapi.security import HTTPBasic, HTTPBasicCredentials
from secrets import compare_digest
import modules.shared as shared
from modules import sd_samplers, deepbooru, sd_hijack
from modules import sd_samplers, deepbooru, sd_hijack, images, scripts, ui
from modules.api.models import *
from modules.processing import StableDiffusionProcessingTxt2Img, StableDiffusionProcessingImg2Img, process_images
from modules.extras import run_extras, run_pnginfo
from modules.extras import run_extras
from modules.textual_inversion.textual_inversion import create_embedding, train_embedding
from modules.textual_inversion.preprocess import preprocess
from modules.hypernetworks.hypernetwork import create_hypernetwork, train_hypernetwork
from PIL import PngImagePlugin,Image
from modules.sd_models import checkpoints_list
from modules.sd_models import checkpoints_list, find_checkpoint_config
from modules.realesrgan_model import get_realesrgan_models
from modules import devices
from typing import List
@ -27,8 +28,13 @@ def upscaler_to_index(name: str):
try:
return [x.name.lower() for x in shared.sd_upscalers].index(name.lower())
except:
raise HTTPException(status_code=400, detail=f"Invalid upscaler, needs to be on of these: {' , '.join([x.name for x in sd_upscalers])}")
raise HTTPException(status_code=400, detail=f"Invalid upscaler, needs to be one of these: {' , '.join([x.name for x in sd_upscalers])}")
def script_name_to_index(name, scripts):
try:
return [script.title().lower() for script in scripts].index(name.lower())
except:
raise HTTPException(status_code=422, detail=f"Script '{name}' not found")
def validate_sampler_name(name):
config = sd_samplers.all_samplers_map.get(name, None)
@ -67,6 +73,27 @@ def encode_pil_to_base64(image):
bytes_data = output_bytes.getvalue()
return base64.b64encode(bytes_data)
def api_middleware(app: FastAPI):
@app.middleware("http")
async def log_and_time(req: Request, call_next):
ts = time.time()
res: Response = await call_next(req)
duration = str(round(time.time() - ts, 4))
res.headers["X-Process-Time"] = duration
endpoint = req.scope.get('path', 'err')
if shared.cmd_opts.api_log and endpoint.startswith('/sdapi'):
print('API {t} {code} {prot}/{ver} {method} {endpoint} {cli} {duration}'.format(
t = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S.%f"),
code = res.status_code,
ver = req.scope.get('http_version', '0.0'),
cli = req.scope.get('client', ('0:0.0.0', 0))[0],
prot = req.scope.get('scheme', 'err'),
method = req.scope.get('method', 'err'),
endpoint = endpoint,
duration = duration,
))
return res
class Api:
def __init__(self, app: FastAPI, queue_lock: Lock):
@ -79,6 +106,7 @@ class Api:
self.router = APIRouter()
self.app = app
self.queue_lock = queue_lock
api_middleware(self.app)
self.add_api_route("/sdapi/v1/txt2img", self.text2imgapi, methods=["POST"], response_model=TextToImageResponse)
self.add_api_route("/sdapi/v1/img2img", self.img2imgapi, methods=["POST"], response_model=ImageToImageResponse)
self.add_api_route("/sdapi/v1/extra-single-image", self.extras_single_image_api, methods=["POST"], response_model=ExtrasSingleImageResponse)
@ -100,12 +128,14 @@ class Api:
self.add_api_route("/sdapi/v1/prompt-styles", self.get_prompt_styles, methods=["GET"], response_model=List[PromptStyleItem])
self.add_api_route("/sdapi/v1/artist-categories", self.get_artists_categories, methods=["GET"], response_model=List[str])
self.add_api_route("/sdapi/v1/artists", self.get_artists, methods=["GET"], response_model=List[ArtistItem])
self.add_api_route("/sdapi/v1/embeddings", self.get_embeddings, methods=["GET"], response_model=EmbeddingsResponse)
self.add_api_route("/sdapi/v1/refresh-checkpoints", self.refresh_checkpoints, methods=["POST"])
self.add_api_route("/sdapi/v1/create/embedding", self.create_embedding, methods=["POST"], response_model=CreateResponse)
self.add_api_route("/sdapi/v1/create/hypernetwork", self.create_hypernetwork, methods=["POST"], response_model=CreateResponse)
self.add_api_route("/sdapi/v1/preprocess", self.preprocess, methods=["POST"], response_model=PreprocessResponse)
self.add_api_route("/sdapi/v1/train/embedding", self.train_embedding, methods=["POST"], response_model=TrainResponse)
self.add_api_route("/sdapi/v1/train/hypernetwork", self.train_hypernetwork, methods=["POST"], response_model=TrainResponse)
self.add_api_route("/sdapi/v1/memory", self.get_memory, methods=["GET"], response_model=MemoryResponse)
def add_api_route(self, path: str, endpoint, **kwargs):
if shared.cmd_opts.api_auth:
@ -119,7 +149,21 @@ class Api:
raise HTTPException(status_code=401, detail="Incorrect username or password", headers={"WWW-Authenticate": "Basic"})
def get_script(self, script_name, script_runner):
if script_name is None:
return None, None
if not script_runner.scripts:
script_runner.initialize_scripts(False)
ui.create_ui()
script_idx = script_name_to_index(script_name, script_runner.selectable_scripts)
script = script_runner.selectable_scripts[script_idx]
return script, script_idx
def text2imgapi(self, txt2imgreq: StableDiffusionTxt2ImgProcessingAPI):
script, script_idx = self.get_script(txt2imgreq.script_name, scripts.scripts_txt2img)
populate = txt2imgreq.copy(update={ # Override __init__ params
"sampler_name": validate_sampler_name(txt2imgreq.sampler_name or txt2imgreq.sampler_index),
"do_not_save_samples": True,
@ -129,14 +173,22 @@ class Api:
if populate.sampler_name:
populate.sampler_index = None # prevent a warning later on
args = vars(populate)
args.pop('script_name', None)
with self.queue_lock:
p = StableDiffusionProcessingTxt2Img(sd_model=shared.sd_model, **vars(populate))
p = StableDiffusionProcessingTxt2Img(sd_model=shared.sd_model, **args)
shared.state.begin()
processed = process_images(p)
if script is not None:
p.outpath_grids = opts.outdir_txt2img_grids
p.outpath_samples = opts.outdir_txt2img_samples
p.script_args = [script_idx + 1] + [None] * (script.args_from - 1) + p.script_args
processed = scripts.scripts_txt2img.run(p, *p.script_args)
else:
processed = process_images(p)
shared.state.end()
b64images = list(map(encode_pil_to_base64, processed.images))
return TextToImageResponse(images=b64images, parameters=vars(txt2imgreq), info=processed.js())
@ -146,6 +198,8 @@ class Api:
if init_images is None:
raise HTTPException(status_code=404, detail="Init image not found")
script, script_idx = self.get_script(img2imgreq.script_name, scripts.scripts_img2img)
mask = img2imgreq.mask
if mask:
mask = decode_base64_to_image(mask)
@ -162,13 +216,20 @@ class Api:
args = vars(populate)
args.pop('include_init_images', None) # this is meant to be done by "exclude": True in model, but it's for a reason that I cannot determine.
args.pop('script_name', None)
with self.queue_lock:
p = StableDiffusionProcessingImg2Img(sd_model=shared.sd_model, **args)
p.init_images = [decode_base64_to_image(x) for x in init_images]
shared.state.begin()
processed = process_images(p)
if script is not None:
p.outpath_grids = opts.outdir_img2img_grids
p.outpath_samples = opts.outdir_img2img_samples
p.script_args = [script_idx + 1] + [None] * (script.args_from - 1) + p.script_args
processed = scripts.scripts_img2img.run(p, *p.script_args)
else:
processed = process_images(p)
shared.state.end()
b64images = list(map(encode_pil_to_base64, processed.images))
@ -209,9 +270,17 @@ class Api:
if(not req.image.strip()):
return PNGInfoResponse(info="")
result = run_pnginfo(decode_base64_to_image(req.image.strip()))
image = decode_base64_to_image(req.image.strip())
if image is None:
return PNGInfoResponse(info="")
return PNGInfoResponse(info=result[1])
geninfo, items = images.read_info_from_image(image)
if geninfo is None:
geninfo = ""
items = {**{'parameters': geninfo}, **items}
return PNGInfoResponse(info=geninfo, items=items)
def progressapi(self, req: ProgressRequest = Depends()):
# copy from check_progress_call of ui.py
@ -302,7 +371,7 @@ class Api:
return upscalers
def get_sd_models(self):
return [{"title":x.title, "model_name":x.model_name, "hash":x.hash, "filename": x.filename, "config": x.config} for x in checkpoints_list.values()]
return [{"title":x.title, "model_name":x.model_name, "hash":x.hash, "filename": x.filename, "config": find_checkpoint_config(x)} for x in checkpoints_list.values()]
def get_hypernetworks(self):
return [{"name": name, "path": shared.hypernetworks[name]} for name in shared.hypernetworks]
@ -327,6 +396,26 @@ class Api:
def get_artists(self):
return [{"name":x[0], "score":x[1], "category":x[2]} for x in shared.artist_db.artists]
def get_embeddings(self):
db = sd_hijack.model_hijack.embedding_db
def convert_embedding(embedding):
return {
"step": embedding.step,
"sd_checkpoint": embedding.sd_checkpoint,
"sd_checkpoint_name": embedding.sd_checkpoint_name,
"shape": embedding.shape,
"vectors": embedding.vectors,
}
def convert_embeddings(embeddings):
return {embedding.name: convert_embedding(embedding) for embedding in embeddings.values()}
return {
"loaded": convert_embeddings(db.word_embeddings),
"skipped": convert_embeddings(db.skipped_embeddings),
}
def refresh_checkpoints(self):
shared.refresh_checkpoints()
@ -413,6 +502,40 @@ class Api:
shared.state.end()
return TrainResponse(info = "train embedding error: {error}".format(error = error))
def get_memory(self):
try:
import os, psutil
process = psutil.Process(os.getpid())
res = process.memory_info() # only rss is cross-platform guaranteed so we dont rely on other values
ram_total = 100 * res.rss / process.memory_percent() # and total memory is calculated as actual value is not cross-platform safe
ram = { 'free': ram_total - res.rss, 'used': res.rss, 'total': ram_total }
except Exception as err:
ram = { 'error': f'{err}' }
try:
import torch
if torch.cuda.is_available():
s = torch.cuda.mem_get_info()
system = { 'free': s[0], 'used': s[1] - s[0], 'total': s[1] }
s = dict(torch.cuda.memory_stats(shared.device))
allocated = { 'current': s['allocated_bytes.all.current'], 'peak': s['allocated_bytes.all.peak'] }
reserved = { 'current': s['reserved_bytes.all.current'], 'peak': s['reserved_bytes.all.peak'] }
active = { 'current': s['active_bytes.all.current'], 'peak': s['active_bytes.all.peak'] }
inactive = { 'current': s['inactive_split_bytes.all.current'], 'peak': s['inactive_split_bytes.all.peak'] }
warnings = { 'retries': s['num_alloc_retries'], 'oom': s['num_ooms'] }
cuda = {
'system': system,
'active': active,
'allocated': allocated,
'reserved': reserved,
'inactive': inactive,
'events': warnings,
}
else:
cuda = { 'error': 'unavailable' }
except Exception as err:
cuda = { 'error': f'{err}' }
return MemoryResponse(ram = ram, cuda = cuda)
def launch(self, server_name, port):
self.app.include_router(self.router)
uvicorn.run(self.app, host=server_name, port=port)

23
modules/api/models.py
View File

@ -100,13 +100,13 @@ class PydanticModelGenerator:
StableDiffusionTxt2ImgProcessingAPI = PydanticModelGenerator(
"StableDiffusionProcessingTxt2Img",
StableDiffusionProcessingTxt2Img,
[{"key": "sampler_index", "type": str, "default": "Euler"}]
[{"key": "sampler_index", "type": str, "default": "Euler"}, {"key": "script_name", "type": str, "default": None}, {"key": "script_args", "type": list, "default": []}]
).generate_model()
StableDiffusionImg2ImgProcessingAPI = PydanticModelGenerator(
"StableDiffusionProcessingImg2Img",
StableDiffusionProcessingImg2Img,
[{"key": "sampler_index", "type": str, "default": "Euler"}, {"key": "init_images", "type": list, "default": None}, {"key": "denoising_strength", "type": float, "default": 0.75}, {"key": "mask", "type": str, "default": None}, {"key": "include_init_images", "type": bool, "default": False, "exclude" : True}]
[{"key": "sampler_index", "type": str, "default": "Euler"}, {"key": "init_images", "type": list, "default": None}, {"key": "denoising_strength", "type": float, "default": 0.75}, {"key": "mask", "type": str, "default": None}, {"key": "include_init_images", "type": bool, "default": False, "exclude" : True}, {"key": "script_name", "type": str, "default": None}, {"key": "script_args", "type": list, "default": []}]
).generate_model()
class TextToImageResponse(BaseModel):
@ -125,7 +125,7 @@ class ExtrasBaseRequest(BaseModel):
gfpgan_visibility: float = Field(default=0, title="GFPGAN Visibility", ge=0, le=1, allow_inf_nan=False, description="Sets the visibility of GFPGAN, values should be between 0 and 1.")
codeformer_visibility: float = Field(default=0, title="CodeFormer Visibility", ge=0, le=1, allow_inf_nan=False, description="Sets the visibility of CodeFormer, values should be between 0 and 1.")
codeformer_weight: float = Field(default=0, title="CodeFormer Weight", ge=0, le=1, allow_inf_nan=False, description="Sets the weight of CodeFormer, values should be between 0 and 1.")
upscaling_resize: float = Field(default=2, title="Upscaling Factor", ge=1, le=4, description="By how much to upscale the image, only used when resize_mode=0.")
upscaling_resize: float = Field(default=2, title="Upscaling Factor", ge=1, le=8, description="By how much to upscale the image, only used when resize_mode=0.")
upscaling_resize_w: int = Field(default=512, title="Target Width", ge=1, description="Target width for the upscaler to hit. Only used when resize_mode=1.")
upscaling_resize_h: int = Field(default=512, title="Target Height", ge=1, description="Target height for the upscaler to hit. Only used when resize_mode=1.")
upscaling_crop: bool = Field(default=True, title="Crop to fit", description="Should the upscaler crop the image to fit in the chosen size?")
@ -157,7 +157,8 @@ class PNGInfoRequest(BaseModel):
image: str = Field(title="Image", description="The base64 encoded PNG image")
class PNGInfoResponse(BaseModel):
info: str = Field(title="Image info", description="A string with all the info the image had")
info: str = Field(title="Image info", description="A string with the parameters used to generate the image")
items: dict = Field(title="Items", description="An object containing all the info the image had")
class ProgressRequest(BaseModel):
skip_current_image: bool = Field(default=False, title="Skip current image", description="Skip current image serialization")
@ -249,3 +250,17 @@ class ArtistItem(BaseModel):
score: float = Field(title="Score")
category: str = Field(title="Category")
class EmbeddingItem(BaseModel):
step: Optional[int] = Field(title="Step", description="The number of steps that were used to train this embedding, if available")
sd_checkpoint: Optional[str] = Field(title="SD Checkpoint", description="The hash of the checkpoint this embedding was trained on, if available")
sd_checkpoint_name: Optional[str] = Field(title="SD Checkpoint Name", description="The name of the checkpoint this embedding was trained on, if available. Note that this is the name that was used by the trainer; for a stable identifier, use `sd_checkpoint` instead")
shape: int = Field(title="Shape", description="The length of each individual vector in the embedding")
vectors: int = Field(title="Vectors", description="The number of vectors in the embedding")
class EmbeddingsResponse(BaseModel):
loaded: Dict[str, EmbeddingItem] = Field(title="Loaded", description="Embeddings loaded for the current model")
skipped: Dict[str, EmbeddingItem] = Field(title="Skipped", description="Embeddings skipped for the current model (likely due to architecture incompatibility)")
class MemoryResponse(BaseModel):
ram: dict = Field(title="RAM", description="System memory stats")
cuda: dict = Field(title="CUDA", description="nVidia CUDA memory stats")

28
modules/devices.py
View File

@ -133,8 +133,26 @@ def numpy_fix(self, *args, **kwargs):
return orig_tensor_numpy(self, *args, **kwargs)
# PyTorch 1.13 doesn't need these fixes but unfortunately is slower and has regressions that prevent training from working
if has_mps() and version.parse(torch.__version__) < version.parse("1.13"):
torch.Tensor.to = tensor_to_fix
torch.nn.functional.layer_norm = layer_norm_fix
torch.Tensor.numpy = numpy_fix
# MPS workaround for https://github.com/pytorch/pytorch/issues/89784
orig_cumsum = torch.cumsum
orig_Tensor_cumsum = torch.Tensor.cumsum
def cumsum_fix(input, cumsum_func, *args, **kwargs):
if input.device.type == 'mps':
output_dtype = kwargs.get('dtype', input.dtype)
if any(output_dtype == broken_dtype for broken_dtype in [torch.bool, torch.int8, torch.int16, torch.int64]):
return cumsum_func(input.cpu(), *args, **kwargs).to(input.device)
return cumsum_func(input, *args, **kwargs)
if has_mps():
if version.parse(torch.__version__) < version.parse("1.13"):
# PyTorch 1.13 doesn't need these fixes but unfortunately is slower and has regressions that prevent training from working
torch.Tensor.to = tensor_to_fix
torch.nn.functional.layer_norm = layer_norm_fix
torch.Tensor.numpy = numpy_fix
elif version.parse(torch.__version__) > version.parse("1.13.1"):
if not torch.Tensor([1,2]).to(torch.device("mps")).equal(torch.Tensor([1,1]).to(torch.device("mps")).cumsum(0, dtype=torch.int16)):
torch.cumsum = lambda input, *args, **kwargs: ( cumsum_fix(input, orig_cumsum, *args, **kwargs) )
torch.Tensor.cumsum = lambda self, *args, **kwargs: ( cumsum_fix(self, orig_Tensor_cumsum, *args, **kwargs) )
orig_narrow = torch.narrow
torch.narrow = lambda *args, **kwargs: ( orig_narrow(*args, **kwargs).clone() )

25
modules/errors.py
View File

@ -2,9 +2,30 @@ import sys
import traceback
def print_error_explanation(message):
lines = message.strip().split("\n")
max_len = max([len(x) for x in lines])
print('=' * max_len, file=sys.stderr)
for line in lines:
print(line, file=sys.stderr)
print('=' * max_len, file=sys.stderr)
def display(e: Exception, task):
print(f"{task or 'error'}: {type(e).__name__}", file=sys.stderr)
print(traceback.format_exc(), file=sys.stderr)
message = str(e)
if "copying a param with shape torch.Size([640, 1024]) from checkpoint, the shape in current model is torch.Size([640, 768])" in message:
print_error_explanation("""
The most likely cause of this is you are trying to load Stable Diffusion 2.0 model without specifying its connfig file.
See https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#stable-diffusion-20 for how to solve this.
""")
def run(code, task):
try:
code()
except Exception as e:
print(f"{task}: {type(e).__name__}", file=sys.stderr)
print(traceback.format_exc(), file=sys.stderr)
display(task, e)

31
modules/extras.py
View File

@ -19,8 +19,6 @@ from modules.shared import opts
import modules.gfpgan_model
from modules.ui import plaintext_to_html
import modules.codeformer_model
import piexif
import piexif.helper
import gradio as gr
import safetensors.torch
@ -58,6 +56,9 @@ cached_images: LruCache = LruCache(max_size=5)
def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_dir, show_extras_results, gfpgan_visibility, codeformer_visibility, codeformer_weight, upscaling_resize, upscaling_resize_w, upscaling_resize_h, upscaling_crop, extras_upscaler_1, extras_upscaler_2, extras_upscaler_2_visibility, upscale_first: bool, save_output: bool = True):
devices.torch_gc()
shared.state.begin()
shared.state.job = 'extras'
imageArr = []
# Also keep track of original file names
imageNameArr = []
@ -94,6 +95,7 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_
# Extra operation definitions
def run_gfpgan(image: Image.Image, info: str) -> Tuple[Image.Image, str]:
shared.state.job = 'extras-gfpgan'
restored_img = modules.gfpgan_model.gfpgan_fix_faces(np.array(image, dtype=np.uint8))
res = Image.fromarray(restored_img)
@ -104,6 +106,7 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_
return (res, info)
def run_codeformer(image: Image.Image, info: str) -> Tuple[Image.Image, str]:
shared.state.job = 'extras-codeformer'
restored_img = modules.codeformer_model.codeformer.restore(np.array(image, dtype=np.uint8), w=codeformer_weight)
res = Image.fromarray(restored_img)
@ -114,6 +117,7 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_
return (res, info)
def upscale(image, scaler_index, resize, mode, resize_w, resize_h, crop):
shared.state.job = 'extras-upscale'
upscaler = shared.sd_upscalers[scaler_index]
res = upscaler.scaler.upscale(image, resize, upscaler.data_path)
if mode == 1 and crop:
@ -180,6 +184,9 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_
for image, image_name in zip(imageArr, imageNameArr):
if image is None:
return outputs, "Please select an input image.", ''
shared.state.textinfo = f'Processing image {image_name}'
existing_pnginfo = image.info or {}
image = image.convert("RGB")
@ -193,6 +200,10 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_
else:
basename = ''
if opts.enable_pnginfo: # append info before save
image.info = existing_pnginfo
image.info["extras"] = info
if save_output:
# Add upscaler name as a suffix.
suffix = f"-{shared.sd_upscalers[extras_upscaler_1].name}" if shared.opts.use_upscaler_name_as_suffix else ""
@ -203,10 +214,6 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_
images.save_image(image, path=outpath, basename=basename, seed=None, prompt=None, extension=opts.samples_format, info=info, short_filename=True,
no_prompt=True, grid=False, pnginfo_section_name="extras", existing_info=existing_pnginfo, forced_filename=None, suffix=suffix)
if opts.enable_pnginfo:
image.info = existing_pnginfo
image.info["extras"] = info
if extras_mode != 2 or show_extras_results :
outputs.append(image)
@ -242,6 +249,9 @@ def run_pnginfo(image):
def run_modelmerger(primary_model_name, secondary_model_name, tertiary_model_name, interp_method, multiplier, save_as_half, custom_name, checkpoint_format):
shared.state.begin()
shared.state.job = 'model-merge'
def weighted_sum(theta0, theta1, alpha):
return ((1 - alpha) * theta0) + (alpha * theta1)
@ -263,8 +273,11 @@ def run_modelmerger(primary_model_name, secondary_model_name, tertiary_model_nam
theta_func1, theta_func2 = theta_funcs[interp_method]
if theta_func1 and not tertiary_model_info:
shared.state.textinfo = "Failed: Interpolation method requires a tertiary model."
shared.state.end()
return ["Failed: Interpolation method requires a tertiary model."] + [gr.Dropdown.update(choices=sd_models.checkpoint_tiles()) for _ in range(4)]
shared.state.textinfo = f"Loading {secondary_model_info.filename}..."
print(f"Loading {secondary_model_info.filename}...")
theta_1 = sd_models.read_state_dict(secondary_model_info.filename, map_location='cpu')
@ -281,6 +294,7 @@ def run_modelmerger(primary_model_name, secondary_model_name, tertiary_model_nam
theta_1[key] = torch.zeros_like(theta_1[key])
del theta_2
shared.state.textinfo = f"Loading {primary_model_info.filename}..."
print(f"Loading {primary_model_info.filename}...")
theta_0 = sd_models.read_state_dict(primary_model_info.filename, map_location='cpu')
@ -291,6 +305,7 @@ def run_modelmerger(primary_model_name, secondary_model_name, tertiary_model_nam
a = theta_0[key]
b = theta_1[key]
shared.state.textinfo = f'Merging layer {key}'
# this enables merging an inpainting model (A) with another one (B);
# where normal model would have 4 channels, for latenst space, inpainting model would
# have another 4 channels for unmasked picture's latent space, plus one channel for mask, for a total of 9
@ -330,6 +345,7 @@ def run_modelmerger(primary_model_name, secondary_model_name, tertiary_model_nam
output_modelname = os.path.join(ckpt_dir, filename)
shared.state.textinfo = f"Saving to {output_modelname}..."
print(f"Saving to {output_modelname}...")
_, extension = os.path.splitext(output_modelname)
@ -341,4 +357,7 @@ def run_modelmerger(primary_model_name, secondary_model_name, tertiary_model_nam
sd_models.list_models()
print("Checkpoint saved.")
shared.state.textinfo = "Checkpoint saved to " + output_modelname
shared.state.end()
return ["Checkpoint saved to " + output_modelname] + [gr.Dropdown.update(choices=sd_models.checkpoint_tiles()) for _ in range(4)]

111
modules/generation_parameters_copypaste.py
View File

@ -1,12 +1,13 @@
import base64
import io
import math
import os
import re
from pathlib import Path
import gradio as gr
from modules.shared import script_path
from modules import shared
from modules import shared, ui_tempdir
import tempfile
from PIL import Image
@ -36,9 +37,12 @@ def quote(text):
def image_from_url_text(filedata):
if type(filedata) == dict and filedata["is_file"]:
if type(filedata) == list and len(filedata) > 0 and type(filedata[0]) == dict and filedata[0].get("is_file", False):
filedata = filedata[0]
if type(filedata) == dict and filedata.get("is_file", False):
filename = filedata["name"]
is_in_right_dir = any([filename in fileset for fileset in shared.demo.temp_file_sets])
is_in_right_dir = ui_tempdir.check_tmp_file(shared.demo, filename)
assert is_in_right_dir, 'trying to open image file outside of allowed directories'
return Image.open(filename)
@ -93,7 +97,7 @@ def integrate_settings_paste_fields(component_dict):
def create_buttons(tabs_list):
buttons = {}
for tab in tabs_list:
buttons[tab] = gr.Button(f"Send to {tab}")
buttons[tab] = gr.Button(f"Send to {tab}", elem_id=f"{tab}_tab")
return buttons
@ -102,35 +106,57 @@ def bind_buttons(buttons, send_image, send_generate_info):
bind_list.append([buttons, send_image, send_generate_info])
def send_image_and_dimensions(x):
if isinstance(x, Image.Image):
img = x
else:
img = image_from_url_text(x)
if shared.opts.send_size and isinstance(img, Image.Image):
w = img.width
h = img.height
else:
w = gr.update()
h = gr.update()
return img, w, h
def run_bind():
for buttons, send_image, send_generate_info in bind_list:
for buttons, source_image_component, send_generate_info in bind_list:
for tab in buttons:
button = buttons[tab]
if send_image and paste_fields[tab]["init_img"]:
if type(send_image) == gr.Gallery:
button.click(
fn=lambda x: image_from_url_text(x),
_js="extract_image_from_gallery",
inputs=[send_image],
outputs=[paste_fields[tab]["init_img"]],
)
else:
button.click(
fn=lambda x: x,
inputs=[send_image],
outputs=[paste_fields[tab]["init_img"]],
)
destination_image_component = paste_fields[tab]["init_img"]
fields = paste_fields[tab]["fields"]
if send_generate_info and paste_fields[tab]["fields"] is not None:
destination_width_component = next(iter([field for field, name in fields if name == "Size-1"] if fields else []), None)
destination_height_component = next(iter([field for field, name in fields if name == "Size-2"] if fields else []), None)
if source_image_component and destination_image_component:
if isinstance(source_image_component, gr.Gallery):
func = send_image_and_dimensions if destination_width_component else image_from_url_text
jsfunc = "extract_image_from_gallery"
else:
func = send_image_and_dimensions if destination_width_component else lambda x: x
jsfunc = None
button.click(
fn=func,
_js=jsfunc,
inputs=[source_image_component],
outputs=[destination_image_component, destination_width_component, destination_height_component] if destination_width_component else [destination_image_component],
)
if send_generate_info and fields is not None:
if send_generate_info in paste_fields:
paste_field_names = ['Prompt', 'Negative prompt', 'Steps', 'Face restoration'] + (['Size-1', 'Size-2'] if shared.opts.send_size else []) + (["Seed"] if shared.opts.send_seed else [])
paste_field_names = ['Prompt', 'Negative prompt', 'Steps', 'Face restoration'] + (["Seed"] if shared.opts.send_seed else [])
button.click(
fn=lambda *x: x,
inputs=[field for field, name in paste_fields[send_generate_info]["fields"] if name in paste_field_names],
outputs=[field for field, name in paste_fields[tab]["fields"] if name in paste_field_names],
outputs=[field for field, name in fields if name in paste_field_names],
)
else:
connect_paste(button, paste_fields[tab]["fields"], send_generate_info)
connect_paste(button, fields, send_generate_info)
button.click(
fn=None,
@ -164,6 +190,39 @@ def find_hypernetwork_key(hypernet_name, hypernet_hash=None):
return None
def restore_old_hires_fix_params(res):
"""for infotexts that specify old First pass size parameter, convert it into
width, height, and hr scale"""
firstpass_width = res.get('First pass size-1', None)
firstpass_height = res.get('First pass size-2', None)
if shared.opts.use_old_hires_fix_width_height:
hires_width = int(res.get("Hires resize-1", 0))
hires_height = int(res.get("Hires resize-2", 0))
if hires_width and hires_height:
res['Size-1'] = hires_width
res['Size-2'] = hires_height
return
if firstpass_width is None or firstpass_height is None:
return
firstpass_width, firstpass_height = int(firstpass_width), int(firstpass_height)
width = int(res.get("Size-1", 512))
height = int(res.get("Size-2", 512))
if firstpass_width == 0 or firstpass_height == 0:
from modules import processing
firstpass_width, firstpass_height = processing.old_hires_fix_first_pass_dimensions(width, height)
res['Size-1'] = firstpass_width
res['Size-2'] = firstpass_height
res['Hires resize-1'] = width
res['Hires resize-2'] = height
def parse_generation_parameters(x: str):
"""parses generation parameters string, the one you see in text field under the picture in UI:
```
@ -221,6 +280,12 @@ Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 965400086, Size: 512x512, Model
hypernet_hash = res.get("Hypernet hash", None)
res["Hypernet"] = find_hypernetwork_key(hypernet_name, hypernet_hash)
if "Hires resize-1" not in res:
res["Hires resize-1"] = 0
res["Hires resize-2"] = 0
restore_old_hires_fix_params(res)
return res

143
modules/hypernetworks/hypernetwork.py
View File

@ -13,7 +13,7 @@ import tqdm
from einops import rearrange, repeat
from ldm.util import default
from modules import devices, processing, sd_models, shared, sd_samplers
from modules.textual_inversion import textual_inversion
from modules.textual_inversion import textual_inversion, logging
from modules.textual_inversion.learn_schedule import LearnRateScheduler
from torch import einsum
from torch.nn.init import normal_, xavier_normal_, xavier_uniform_, kaiming_normal_, kaiming_uniform_, zeros_
@ -24,6 +24,7 @@ from statistics import stdev, mean
optimizer_dict = {optim_name : cls_obj for optim_name, cls_obj in inspect.getmembers(torch.optim, inspect.isclass) if optim_name != "Optimizer"}
class HypernetworkModule(torch.nn.Module):
multiplier = 1.0
activation_dict = {
@ -38,7 +39,7 @@ class HypernetworkModule(torch.nn.Module):
activation_dict.update({cls_name.lower(): cls_obj for cls_name, cls_obj in inspect.getmembers(torch.nn.modules.activation) if inspect.isclass(cls_obj) and cls_obj.__module__ == 'torch.nn.modules.activation'})
def __init__(self, dim, state_dict=None, layer_structure=None, activation_func=None, weight_init='Normal',
add_layer_norm=False, use_dropout=False, activate_output=False, last_layer_dropout=False):
add_layer_norm=False, activate_output=False, dropout_structure=None):
super().__init__()
assert layer_structure is not None, "layer_structure must not be None"
@ -63,9 +64,12 @@ class HypernetworkModule(torch.nn.Module):
if add_layer_norm:
linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1])))
# Add dropout except last layer
if use_dropout and (i < len(layer_structure) - 3 or last_layer_dropout and i < len(layer_structure) - 2):
linears.append(torch.nn.Dropout(p=0.3))
# Everything should be now parsed into dropout structure, and applied here.
# Since we only have dropouts after layers, dropout structure should start with 0 and end with 0.
if dropout_structure is not None and dropout_structure[i+1] > 0:
assert 0 < dropout_structure[i+1] < 1, "Dropout probability should be 0 or float between 0 and 1!"
linears.append(torch.nn.Dropout(p=dropout_structure[i+1]))
# Code explanation : [1, 2, 1] -> dropout is missing when last_layer_dropout is false. [1, 2, 2, 1] -> [0, 0.3, 0, 0], when its True, [0, 0.3, 0.3, 0].
self.linear = torch.nn.Sequential(*linears)
@ -112,7 +116,7 @@ class HypernetworkModule(torch.nn.Module):
state_dict[to] = x
def forward(self, x):
return x + self.linear(x) * self.multiplier
return x + self.linear(x) * (HypernetworkModule.multiplier if not self.training else 1)
def trainables(self):
layer_structure = []
@ -125,6 +129,21 @@ class HypernetworkModule(torch.nn.Module):
def apply_strength(value=None):
HypernetworkModule.multiplier = value if value is not None else shared.opts.sd_hypernetwork_strength
#param layer_structure : sequence used for length, use_dropout : controlling boolean, last_layer_dropout : for compatibility check.
def parse_dropout_structure(layer_structure, use_dropout, last_layer_dropout):
if layer_structure is None:
layer_structure = [1, 2, 1]
if not use_dropout:
return [0] * len(layer_structure)
dropout_values = [0]
dropout_values.extend([0.3] * (len(layer_structure) - 3))
if last_layer_dropout:
dropout_values.append(0.3)
else:
dropout_values.append(0)
dropout_values.append(0)
return dropout_values
class Hypernetwork:
filename = None
@ -143,18 +162,22 @@ class Hypernetwork:
self.add_layer_norm = add_layer_norm
self.use_dropout = use_dropout
self.activate_output = activate_output
self.last_layer_dropout = kwargs['last_layer_dropout'] if 'last_layer_dropout' in kwargs else True
self.last_layer_dropout = kwargs.get('last_layer_dropout', True)
self.dropout_structure = kwargs.get('dropout_structure', None)
if self.dropout_structure is None:
self.dropout_structure = parse_dropout_structure(self.layer_structure, self.use_dropout, self.last_layer_dropout)
self.optimizer_name = None
self.optimizer_state_dict = None
self.optional_info = None
for size in enable_sizes or []:
self.layers[size] = (
HypernetworkModule(size, None, self.layer_structure, self.activation_func, self.weight_init,
self.add_layer_norm, self.use_dropout, self.activate_output, last_layer_dropout=self.last_layer_dropout),
self.add_layer_norm, self.activate_output, dropout_structure=self.dropout_structure),
HypernetworkModule(size, None, self.layer_structure, self.activation_func, self.weight_init,
self.add_layer_norm, self.use_dropout, self.activate_output, last_layer_dropout=self.last_layer_dropout),
self.add_layer_norm, self.activate_output, dropout_structure=self.dropout_structure),
)
self.eval_mode()
self.eval()
def weights(self):
res = []
@ -163,14 +186,14 @@ class Hypernetwork:
res += layer.parameters()
return res
def train_mode(self):
def train(self, mode=True):
for k, layers in self.layers.items():
for layer in layers:
layer.train()
layer.train(mode=mode)
for param in layer.parameters():
param.requires_grad = True
param.requires_grad = mode
def eval_mode(self):
def eval(self):
for k, layers in self.layers.items():
for layer in layers:
layer.eval()
@ -190,11 +213,13 @@ class Hypernetwork:
state_dict['activation_func'] = self.activation_func
state_dict['is_layer_norm'] = self.add_layer_norm
state_dict['weight_initialization'] = self.weight_init
state_dict['use_dropout'] = self.use_dropout
state_dict['sd_checkpoint'] = self.sd_checkpoint
state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name
state_dict['activate_output'] = self.activate_output
state_dict['last_layer_dropout'] = self.last_layer_dropout
state_dict['use_dropout'] = self.use_dropout
state_dict['dropout_structure'] = self.dropout_structure
state_dict['last_layer_dropout'] = (self.dropout_structure[-2] != 0) if self.dropout_structure is not None else self.last_layer_dropout
state_dict['optional_info'] = self.optional_info if self.optional_info else None
if self.optimizer_name is not None:
optimizer_saved_dict['optimizer_name'] = self.optimizer_name
@ -214,43 +239,56 @@ class Hypernetwork:
self.layer_structure = state_dict.get('layer_structure', [1, 2, 1])
print(self.layer_structure)
optional_info = state_dict.get('optional_info', None)
if optional_info is not None:
print(f"INFO:\n {optional_info}\n")
self.optional_info = optional_info
self.activation_func = state_dict.get('activation_func', None)
print(f"Activation function is {self.activation_func}")
self.weight_init = state_dict.get('weight_initialization', 'Normal')
print(f"Weight initialization is {self.weight_init}")
self.add_layer_norm = state_dict.get('is_layer_norm', False)
print(f"Layer norm is set to {self.add_layer_norm}")
self.use_dropout = state_dict.get('use_dropout', False)
self.dropout_structure = state_dict.get('dropout_structure', None)
self.use_dropout = True if self.dropout_structure is not None and any(self.dropout_structure) else state_dict.get('use_dropout', False)
print(f"Dropout usage is set to {self.use_dropout}" )
self.activate_output = state_dict.get('activate_output', True)
print(f"Activate last layer is set to {self.activate_output}")
self.last_layer_dropout = state_dict.get('last_layer_dropout', False)
# Dropout structure should have same length as layer structure, Every digits should be in [0,1), and last digit must be 0.
if self.dropout_structure is None:
print("Using previous dropout structure")
self.dropout_structure = parse_dropout_structure(self.layer_structure, self.use_dropout, self.last_layer_dropout)
print(f"Dropout structure is set to {self.dropout_structure}")
optimizer_saved_dict = torch.load(self.filename + '.optim', map_location = 'cpu') if os.path.exists(self.filename + '.optim') else {}
self.optimizer_name = optimizer_saved_dict.get('optimizer_name', 'AdamW')
print(f"Optimizer name is {self.optimizer_name}")
if sd_models.model_hash(filename) == optimizer_saved_dict.get('hash', None):
self.optimizer_state_dict = optimizer_saved_dict.get('optimizer_state_dict', None)
else:
self.optimizer_state_dict = None
if self.optimizer_state_dict:
self.optimizer_name = optimizer_saved_dict.get('optimizer_name', 'AdamW')
print("Loaded existing optimizer from checkpoint")
print(f"Optimizer name is {self.optimizer_name}")
else:
self.optimizer_name = "AdamW"
print("No saved optimizer exists in checkpoint")
for size, sd in state_dict.items():
if type(size) == int:
self.layers[size] = (
HypernetworkModule(size, sd[0], self.layer_structure, self.activation_func, self.weight_init,
self.add_layer_norm, self.use_dropout, self.activate_output, last_layer_dropout=self.last_layer_dropout),
self.add_layer_norm, self.activate_output, self.dropout_structure),
HypernetworkModule(size, sd[1], self.layer_structure, self.activation_func, self.weight_init,
self.add_layer_norm, self.use_dropout, self.activate_output, last_layer_dropout=self.last_layer_dropout),
self.add_layer_norm, self.activate_output, self.dropout_structure),
)
self.name = state_dict.get('name', self.name)
self.step = state_dict.get('step', 0)
self.sd_checkpoint = state_dict.get('sd_checkpoint', None)
self.sd_checkpoint_name = state_dict.get('sd_checkpoint_name', None)
self.eval()
def list_hypernetworks(path):
@ -378,9 +416,10 @@ def report_statistics(loss_info:dict):
print(e)
def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, activation_func=None, weight_init=None, add_layer_norm=False, use_dropout=False):
def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, activation_func=None, weight_init=None, add_layer_norm=False, use_dropout=False, dropout_structure=None):
# Remove illegal characters from name.
name = "".join( x for x in name if (x.isalnum() or x in "._- "))
assert name, "Name cannot be empty!"
fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt")
if not overwrite_old:
@ -389,6 +428,11 @@ def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None,
if type(layer_structure) == str:
layer_structure = [float(x.strip()) for x in layer_structure.split(",")]
if use_dropout and dropout_structure and type(dropout_structure) == str:
dropout_structure = [float(x.strip()) for x in dropout_structure.split(",")]
else:
dropout_structure = [0] * len(layer_structure)
hypernet = modules.hypernetworks.hypernetwork.Hypernetwork(
name=name,
enable_sizes=[int(x) for x in enable_sizes],
@ -397,26 +441,28 @@ def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None,
weight_init=weight_init,
add_layer_norm=add_layer_norm,
use_dropout=use_dropout,
dropout_structure=dropout_structure
)
hypernet.save(fn)
shared.reload_hypernetworks()
return fn
def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step, data_root, log_directory, training_width, training_height, steps, shuffle_tags, tag_drop_out, latent_sampling_method, create_image_every, save_hypernetwork_every, template_file, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step, data_root, log_directory, training_width, training_height, varsize, steps, clip_grad_mode, clip_grad_value, shuffle_tags, tag_drop_out, latent_sampling_method, create_image_every, save_hypernetwork_every, template_filename, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
# images allows training previews to have infotext. Importing it at the top causes a circular import problem.
from modules import images
save_hypernetwork_every = save_hypernetwork_every or 0
create_image_every = create_image_every or 0
textual_inversion.validate_train_inputs(hypernetwork_name, learn_rate, batch_size, gradient_step, data_root, template_file, steps, save_hypernetwork_every, create_image_every, log_directory, name="hypernetwork")
template_file = textual_inversion.textual_inversion_templates.get(template_filename, None)
textual_inversion.validate_train_inputs(hypernetwork_name, learn_rate, batch_size, gradient_step, data_root, template_file, template_filename, steps, save_hypernetwork_every, create_image_every, log_directory, name="hypernetwork")
template_file = template_file.path
path = shared.hypernetworks.get(hypernetwork_name, None)
shared.loaded_hypernetwork = Hypernetwork()
shared.loaded_hypernetwork.load(path)
shared.state.job = "train-hypernetwork"
shared.state.textinfo = "Initializing hypernetwork training..."
shared.state.job_count = steps
@ -447,14 +493,25 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step,
return hypernetwork, filename
scheduler = LearnRateScheduler(learn_rate, steps, initial_step)
clip_grad = torch.nn.utils.clip_grad_value_ if clip_grad_mode == "value" else torch.nn.utils.clip_grad_norm_ if clip_grad_mode == "norm" else None
if clip_grad:
clip_grad_sched = LearnRateScheduler(clip_grad_value, steps, initial_step, verbose=False)
# dataset loading may take a while, so input validations and early returns should be done before this
shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
pin_memory = shared.opts.pin_memory
ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=hypernetwork_name, model=shared.sd_model, cond_model=shared.sd_model.cond_stage_model, device=devices.device, template_file=template_file, include_cond=True, batch_size=batch_size, gradient_step=gradient_step, shuffle_tags=shuffle_tags, tag_drop_out=tag_drop_out, latent_sampling_method=latent_sampling_method)
ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=hypernetwork_name, model=shared.sd_model, cond_model=shared.sd_model.cond_stage_model, device=devices.device, template_file=template_file, include_cond=True, batch_size=batch_size, gradient_step=gradient_step, shuffle_tags=shuffle_tags, tag_drop_out=tag_drop_out, latent_sampling_method=latent_sampling_method, varsize=varsize)
if shared.opts.save_training_settings_to_txt:
saved_params = dict(
model_name=checkpoint.model_name, model_hash=checkpoint.hash, num_of_dataset_images=len(ds),
**{field: getattr(hypernetwork, field) for field in ['layer_structure', 'activation_func', 'weight_init', 'add_layer_norm', 'use_dropout', ]}
)
logging.save_settings_to_file(log_directory, {**saved_params, **locals()})
latent_sampling_method = ds.latent_sampling_method
dl = modules.textual_inversion.dataset.PersonalizedDataLoader(ds, latent_sampling_method=latent_sampling_method, batch_size=ds.batch_size, pin_memory=pin_memory)
@ -465,9 +522,9 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step,
shared.parallel_processing_allowed = False
shared.sd_model.cond_stage_model.to(devices.cpu)
shared.sd_model.first_stage_model.to(devices.cpu)
weights = hypernetwork.weights()
hypernetwork.train_mode()
hypernetwork.train()
# Here we use optimizer from saved HN, or we can specify as UI option.
if hypernetwork.optimizer_name in optimizer_dict:
@ -524,6 +581,9 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step,
if shared.state.interrupted:
break
if clip_grad:
clip_grad_sched.step(hypernetwork.step)
with devices.autocast():
x = batch.latent_sample.to(devices.device, non_blocking=pin_memory)
if tag_drop_out != 0 or shuffle_tags:
@ -538,14 +598,14 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step,
_loss_step += loss.item()
scaler.scale(loss).backward()
# go back until we reach gradient accumulation steps
if (j + 1) % gradient_step != 0:
continue
# print(f"grad:{weights[0].grad.detach().cpu().abs().mean().item():.7f}")
# scaler.unscale_(optimizer)
# print(f"grad:{weights[0].grad.detach().cpu().abs().mean().item():.15f}")
# torch.nn.utils.clip_grad_norm_(weights, max_norm=1.0)
# print(f"grad:{weights[0].grad.detach().cpu().abs().mean().item():.15f}")
if clip_grad:
clip_grad(weights, clip_grad_sched.learn_rate)
scaler.step(optimizer)
scaler.update()
hypernetwork.step += 1
@ -578,7 +638,11 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step,
if images_dir is not None and steps_done % create_image_every == 0:
forced_filename = f'{hypernetwork_name}-{steps_done}'
last_saved_image = os.path.join(images_dir, forced_filename)
hypernetwork.eval_mode()
hypernetwork.eval()
rng_state = torch.get_rng_state()
cuda_rng_state = None
if torch.cuda.is_available():
cuda_rng_state = torch.cuda.get_rng_state_all()
shared.sd_model.cond_stage_model.to(devices.device)
shared.sd_model.first_stage_model.to(devices.device)
@ -611,7 +675,10 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step,
if unload:
shared.sd_model.cond_stage_model.to(devices.cpu)
shared.sd_model.first_stage_model.to(devices.cpu)
hypernetwork.train_mode()
torch.set_rng_state(rng_state)
if torch.cuda.is_available():
torch.cuda.set_rng_state_all(cuda_rng_state)
hypernetwork.train()
if image is not None:
shared.state.current_image = image
last_saved_image, last_text_info = images.save_image(image, images_dir, "", p.seed, p.prompt, shared.opts.samples_format, processed.infotexts[0], p=p, forced_filename=forced_filename, save_to_dirs=False)
@ -633,7 +700,7 @@ Last saved image: {html.escape(last_saved_image)}<br/>
finally:
pbar.leave = False
pbar.close()
hypernetwork.eval_mode()
hypernetwork.eval()
#report_statistics(loss_dict)
filename = os.path.join(shared.cmd_opts.hypernetwork_dir, f'{hypernetwork_name}.pt')

4
modules/hypernetworks/ui.py
View File

@ -9,8 +9,8 @@ from modules import devices, sd_hijack, shared
not_available = ["hardswish", "multiheadattention"]
keys = list(x for x in modules.hypernetworks.hypernetwork.HypernetworkModule.activation_dict.keys() if x not in not_available)
def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, activation_func=None, weight_init=None, add_layer_norm=False, use_dropout=False):
filename = modules.hypernetworks.hypernetwork.create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure, activation_func, weight_init, add_layer_norm, use_dropout)
def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, activation_func=None, weight_init=None, add_layer_norm=False, use_dropout=False, dropout_structure=None):
filename = modules.hypernetworks.hypernetwork.create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure, activation_func, weight_init, add_layer_norm, use_dropout, dropout_structure)
return gr.Dropdown.update(choices=sorted([x for x in shared.hypernetworks.keys()])), f"Created: {filename}", ""

24
modules/images.py
View File

@ -230,16 +230,32 @@ def draw_prompt_matrix(im, width, height, all_prompts):
return draw_grid_annotations(im, width, height, hor_texts, ver_texts)
def resize_image(resize_mode, im, width, height):
def resize_image(resize_mode, im, width, height, upscaler_name=None):
"""
Resizes an image with the specified resize_mode, width, and height.
Args:
resize_mode: The mode to use when resizing the image.
0: Resize the image to the specified width and height.
1: Resize the image to fill the specified width and height, maintaining the aspect ratio, and then center the image within the dimensions, cropping the excess.
2: Resize the image to fit within the specified width and height, maintaining the aspect ratio, and then center the image within the dimensions, filling empty with data from image.
im: The image to resize.
width: The width to resize the image to.
height: The height to resize the image to.
upscaler_name: The name of the upscaler to use. If not provided, defaults to opts.upscaler_for_img2img.
"""
upscaler_name = upscaler_name or opts.upscaler_for_img2img
def resize(im, w, h):
if opts.upscaler_for_img2img is None or opts.upscaler_for_img2img == "None" or im.mode == 'L':
if upscaler_name is None or upscaler_name == "None" or im.mode == 'L':
return im.resize((w, h), resample=LANCZOS)
scale = max(w / im.width, h / im.height)
if scale > 1.0:
upscalers = [x for x in shared.sd_upscalers if x.name == opts.upscaler_for_img2img]
assert len(upscalers) > 0, f"could not find upscaler named {opts.upscaler_for_img2img}"
upscalers = [x for x in shared.sd_upscalers if x.name == upscaler_name]
assert len(upscalers) > 0, f"could not find upscaler named {upscaler_name}"
upscaler = upscalers[0]
im = upscaler.scaler.upscale(im, scale, upscaler.data_path)

4
modules/interrogate.py
View File

@ -136,7 +136,8 @@ class InterrogateModels:
def interrogate(self, pil_image):
res = ""
shared.state.begin()
shared.state.job = 'interrogate'
try:
if shared.cmd_opts.lowvram or shared.cmd_opts.medvram:
@ -177,5 +178,6 @@ class InterrogateModels:
res += "<error>"
self.unload()
shared.state.end()
return res

24
modules/modelloader.py
View File

@ -10,7 +10,7 @@ from modules.upscaler import Upscaler
from modules.paths import script_path, models_path
def load_models(model_path: str, model_url: str = None, command_path: str = None, ext_filter=None, download_name=None) -> list:
def load_models(model_path: str, model_url: str = None, command_path: str = None, ext_filter=None, download_name=None, ext_blacklist=None) -> list:
"""
A one-and done loader to try finding the desired models in specified directories.
@ -45,6 +45,8 @@ def load_models(model_path: str, model_url: str = None, command_path: str = None
full_path = file
if os.path.isdir(full_path):
continue
if ext_blacklist is not None and any([full_path.endswith(x) for x in ext_blacklist]):
continue
if len(ext_filter) != 0:
model_name, extension = os.path.splitext(file)
if extension not in ext_filter:
@ -123,6 +125,23 @@ def move_files(src_path: str, dest_path: str, ext_filter: str = None):
pass
builtin_upscaler_classes = []
forbidden_upscaler_classes = set()
def list_builtin_upscalers():
load_upscalers()
builtin_upscaler_classes.clear()
builtin_upscaler_classes.extend(Upscaler.__subclasses__())
def forbid_loaded_nonbuiltin_upscalers():
for cls in Upscaler.__subclasses__():
if cls not in builtin_upscaler_classes:
forbidden_upscaler_classes.add(cls)
def load_upscalers():
# We can only do this 'magic' method to dynamically load upscalers if they are referenced,
# so we'll try to import any _model.py files before looking in __subclasses__
@ -139,6 +158,9 @@ def load_upscalers():
datas = []
commandline_options = vars(shared.cmd_opts)
for cls in Upscaler.__subclasses__():
if cls in forbidden_upscaler_classes:
continue
name = cls.__name__
cmd_name = f"{name.lower().replace('upscaler', '')}_models_path"
scaler = cls(commandline_options.get(cmd_name, None))

239
modules/processing.py
View File

@ -50,9 +50,9 @@ def apply_color_correction(correction, original_image):
correction,
channel_axis=2
), cv2.COLOR_LAB2RGB).astype("uint8"))
image = blendLayers(image, original_image, BlendType.LUMINOSITY)
return image
@ -76,11 +76,29 @@ def apply_overlay(image, paste_loc, index, overlays):
return image
def txt2img_image_conditioning(sd_model, x, width, height):
if sd_model.model.conditioning_key not in {'hybrid', 'concat'}:
# Dummy zero conditioning if we're not using inpainting model.
# Still takes up a bit of memory, but no encoder call.
# Pretty sure we can just make this a 1x1 image since its not going to be used besides its batch size.
return x.new_zeros(x.shape[0], 5, 1, 1, dtype=x.dtype, device=x.device)
# The "masked-image" in this case will just be all zeros since the entire image is masked.
image_conditioning = torch.zeros(x.shape[0], 3, height, width, device=x.device)
image_conditioning = sd_model.get_first_stage_encoding(sd_model.encode_first_stage(image_conditioning))
# Add the fake full 1s mask to the first dimension.
image_conditioning = torch.nn.functional.pad(image_conditioning, (0, 0, 0, 0, 1, 0), value=1.0)
image_conditioning = image_conditioning.to(x.dtype)
return image_conditioning
class StableDiffusionProcessing():
"""
The first set of paramaters: sd_models -> do_not_reload_embeddings represent the minimum required to create a StableDiffusionProcessing
"""
def __init__(self, sd_model=None, outpath_samples=None, outpath_grids=None, prompt: str = "", styles: List[str] = None, seed: int = -1, subseed: int = -1, subseed_strength: float = 0, seed_resize_from_h: int = -1, seed_resize_from_w: int = -1, seed_enable_extras: bool = True, sampler_name: str = None, batch_size: int = 1, n_iter: int = 1, steps: int = 50, cfg_scale: float = 7.0, width: int = 512, height: int = 512, restore_faces: bool = False, tiling: bool = False, do_not_save_samples: bool = False, do_not_save_grid: bool = False, extra_generation_params: Dict[Any, Any] = None, overlay_images: Any = None, negative_prompt: str = None, eta: float = None, do_not_reload_embeddings: bool = False, denoising_strength: float = 0, ddim_discretize: str = None, s_churn: float = 0.0, s_tmax: float = None, s_tmin: float = 0.0, s_noise: float = 1.0, override_settings: Dict[str, Any] = None, override_settings_restore_afterwards: bool = True, sampler_index: int = None):
def __init__(self, sd_model=None, outpath_samples=None, outpath_grids=None, prompt: str = "", styles: List[str] = None, seed: int = -1, subseed: int = -1, subseed_strength: float = 0, seed_resize_from_h: int = -1, seed_resize_from_w: int = -1, seed_enable_extras: bool = True, sampler_name: str = None, batch_size: int = 1, n_iter: int = 1, steps: int = 50, cfg_scale: float = 7.0, width: int = 512, height: int = 512, restore_faces: bool = False, tiling: bool = False, do_not_save_samples: bool = False, do_not_save_grid: bool = False, extra_generation_params: Dict[Any, Any] = None, overlay_images: Any = None, negative_prompt: str = None, eta: float = None, do_not_reload_embeddings: bool = False, denoising_strength: float = 0, ddim_discretize: str = None, s_churn: float = 0.0, s_tmax: float = None, s_tmin: float = 0.0, s_noise: float = 1.0, override_settings: Dict[str, Any] = None, override_settings_restore_afterwards: bool = True, sampler_index: int = None, script_args: list = None):
if sampler_index is not None:
print("sampler_index argument for StableDiffusionProcessing does not do anything; use sampler_name", file=sys.stderr)
@ -131,33 +149,17 @@ class StableDiffusionProcessing():
self.seed_resize_from_w = 0
self.scripts = None
self.script_args = None
self.script_args = script_args
self.all_prompts = None
self.all_negative_prompts = None
self.all_seeds = None
self.all_subseeds = None
self.iteration = 0
def txt2img_image_conditioning(self, x, width=None, height=None):
if self.sampler.conditioning_key not in {'hybrid', 'concat'}:
# Dummy zero conditioning if we're not using inpainting model.
# Still takes up a bit of memory, but no encoder call.
# Pretty sure we can just make this a 1x1 image since its not going to be used besides its batch size.
return x.new_zeros(x.shape[0], 5, 1, 1)
self.is_using_inpainting_conditioning = self.sd_model.model.conditioning_key in {'hybrid', 'concat'}
self.is_using_inpainting_conditioning = True
height = height or self.height
width = width or self.width
# The "masked-image" in this case will just be all zeros since the entire image is masked.
image_conditioning = torch.zeros(x.shape[0], 3, height, width, device=x.device)
image_conditioning = self.sd_model.get_first_stage_encoding(self.sd_model.encode_first_stage(image_conditioning))
# Add the fake full 1s mask to the first dimension.
image_conditioning = torch.nn.functional.pad(image_conditioning, (0, 0, 0, 0, 1, 0), value=1.0)
image_conditioning = image_conditioning.to(x.dtype)
return image_conditioning
return txt2img_image_conditioning(self.sd_model, x, width or self.width, height or self.height)
def depth2img_image_conditioning(self, source_image):
# Use the AddMiDaS helper to Format our source image to suit the MiDaS model
@ -420,7 +422,7 @@ def fix_seed(p):
p.subseed = get_fixed_seed(p.subseed)
def create_infotext(p, all_prompts, all_seeds, all_subseeds, comments, iteration=0, position_in_batch=0):
def create_infotext(p, all_prompts, all_seeds, all_subseeds, comments=None, iteration=0, position_in_batch=0):
index = position_in_batch + iteration * p.batch_size
clip_skip = getattr(p, 'clip_skip', opts.CLIP_stop_at_last_layers)
@ -464,9 +466,15 @@ def process_images(p: StableDiffusionProcessing) -> Processed:
try:
for k, v in p.override_settings.items():
setattr(opts, k, v)
if k == 'sd_hypernetwork': shared.reload_hypernetworks() # make onchange call for changing hypernet
if k == 'sd_model_checkpoint': sd_models.reload_model_weights() # make onchange call for changing SD model
if k == 'sd_vae': sd_vae.reload_vae_weights() # make onchange call for changing VAE
if k == 'sd_hypernetwork':
shared.reload_hypernetworks() # make onchange call for changing hypernet
if k == 'sd_model_checkpoint':
sd_models.reload_model_weights() # make onchange call for changing SD model
p.sd_model = shared.sd_model
if k == 'sd_vae':
sd_vae.reload_vae_weights() # make onchange call for changing VAE
res = process_images_inner(p)
@ -536,6 +544,29 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
infotexts = []
output_images = []
cached_uc = [None, None]
cached_c = [None, None]
def get_conds_with_caching(function, required_prompts, steps, cache):
"""
Returns the result of calling function(shared.sd_model, required_prompts, steps)
using a cache to store the result if the same arguments have been used before.
cache is an array containing two elements. The first element is a tuple
representing the previously used arguments, or None if no arguments
have been used before. The second element is where the previously
computed result is stored.
"""
if cache[0] is not None and (required_prompts, steps) == cache[0]:
return cache[1]
with devices.autocast():
cache[1] = function(shared.sd_model, required_prompts, steps)
cache[0] = (required_prompts, steps)
return cache[1]
with torch.no_grad(), p.sd_model.ema_scope():
with devices.autocast():
p.init(p.all_prompts, p.all_seeds, p.all_subseeds)
@ -544,6 +575,8 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
state.job_count = p.n_iter
for n in range(p.n_iter):
p.iteration = n
if state.skipped:
state.skipped = False
@ -561,9 +594,8 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
if p.scripts is not None:
p.scripts.process_batch(p, batch_number=n, prompts=prompts, seeds=seeds, subseeds=subseeds)
with devices.autocast():
uc = prompt_parser.get_learned_conditioning(shared.sd_model, negative_prompts, p.steps)
c = prompt_parser.get_multicond_learned_conditioning(shared.sd_model, prompts, p.steps)
uc = get_conds_with_caching(prompt_parser.get_learned_conditioning, negative_prompts, p.steps, cached_uc)
c = get_conds_with_caching(prompt_parser.get_multicond_learned_conditioning, prompts, p.steps, cached_c)
if len(model_hijack.comments) > 0:
for comment in model_hijack.comments:
@ -655,79 +687,138 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
return res
def old_hires_fix_first_pass_dimensions(width, height):
"""old algorithm for auto-calculating first pass size"""
desired_pixel_count = 512 * 512
actual_pixel_count = width * height
scale = math.sqrt(desired_pixel_count / actual_pixel_count)
width = math.ceil(scale * width / 64) * 64
height = math.ceil(scale * height / 64) * 64
return width, height
class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
sampler = None
def __init__(self, enable_hr: bool=False, denoising_strength: float=0.75, firstphase_width: int=0, firstphase_height: int=0, **kwargs):
def __init__(self, enable_hr: bool = False, denoising_strength: float = 0.75, firstphase_width: int = 0, firstphase_height: int = 0, hr_scale: float = 2.0, hr_upscaler: str = None, hr_second_pass_steps: int = 0, hr_resize_x: int = 0, hr_resize_y: int = 0, **kwargs):
super().__init__(**kwargs)
self.enable_hr = enable_hr
self.denoising_strength = denoising_strength
self.firstphase_width = firstphase_width
self.firstphase_height = firstphase_height
self.hr_scale = hr_scale
self.hr_upscaler = hr_upscaler
self.hr_second_pass_steps = hr_second_pass_steps
self.hr_resize_x = hr_resize_x
self.hr_resize_y = hr_resize_y
self.hr_upscale_to_x = hr_resize_x
self.hr_upscale_to_y = hr_resize_y
if firstphase_width != 0 or firstphase_height != 0:
self.hr_upscale_to_x = self.width
self.hr_upscale_to_y = self.height
self.width = firstphase_width
self.height = firstphase_height
self.truncate_x = 0
self.truncate_y = 0
self.applied_old_hires_behavior_to = None
def init(self, all_prompts, all_seeds, all_subseeds):
if self.enable_hr:
if state.job_count == -1:
state.job_count = self.n_iter * 2
if opts.use_old_hires_fix_width_height and self.applied_old_hires_behavior_to != (self.width, self.height):
self.hr_resize_x = self.width
self.hr_resize_y = self.height
self.hr_upscale_to_x = self.width
self.hr_upscale_to_y = self.height
self.width, self.height = old_hires_fix_first_pass_dimensions(self.width, self.height)
self.applied_old_hires_behavior_to = (self.width, self.height)
if self.hr_resize_x == 0 and self.hr_resize_y == 0:
self.extra_generation_params["Hires upscale"] = self.hr_scale
self.hr_upscale_to_x = int(self.width * self.hr_scale)
self.hr_upscale_to_y = int(self.height * self.hr_scale)
else:
state.job_count = state.job_count * 2
self.extra_generation_params["Hires resize"] = f"{self.hr_resize_x}x{self.hr_resize_y}"
self.extra_generation_params["First pass size"] = f"{self.firstphase_width}x{self.firstphase_height}"
if self.firstphase_width == 0 or self.firstphase_height == 0:
desired_pixel_count = 512 * 512
actual_pixel_count = self.width * self.height
scale = math.sqrt(desired_pixel_count / actual_pixel_count)
self.firstphase_width = math.ceil(scale * self.width / 64) * 64
self.firstphase_height = math.ceil(scale * self.height / 64) * 64
firstphase_width_truncated = int(scale * self.width)
firstphase_height_truncated = int(scale * self.height)
else:
width_ratio = self.width / self.firstphase_width
height_ratio = self.height / self.firstphase_height
if width_ratio > height_ratio:
firstphase_width_truncated = self.firstphase_width
firstphase_height_truncated = self.firstphase_width * self.height / self.width
if self.hr_resize_y == 0:
self.hr_upscale_to_x = self.hr_resize_x
self.hr_upscale_to_y = self.hr_resize_x * self.height // self.width
elif self.hr_resize_x == 0:
self.hr_upscale_to_x = self.hr_resize_y * self.width // self.height
self.hr_upscale_to_y = self.hr_resize_y
else:
firstphase_width_truncated = self.firstphase_height * self.width / self.height
firstphase_height_truncated = self.firstphase_height
target_w = self.hr_resize_x
target_h = self.hr_resize_y
src_ratio = self.width / self.height
dst_ratio = self.hr_resize_x / self.hr_resize_y
self.truncate_x = int(self.firstphase_width - firstphase_width_truncated) // opt_f
self.truncate_y = int(self.firstphase_height - firstphase_height_truncated) // opt_f
if src_ratio < dst_ratio:
self.hr_upscale_to_x = self.hr_resize_x
self.hr_upscale_to_y = self.hr_resize_x * self.height // self.width
else:
self.hr_upscale_to_x = self.hr_resize_y * self.width // self.height
self.hr_upscale_to_y = self.hr_resize_y
self.truncate_x = (self.hr_upscale_to_x - target_w) // opt_f
self.truncate_y = (self.hr_upscale_to_y - target_h) // opt_f
# special case: the user has chosen to do nothing
if self.hr_upscale_to_x == self.width and self.hr_upscale_to_y == self.height:
self.enable_hr = False
self.denoising_strength = None
self.extra_generation_params.pop("Hires upscale", None)
self.extra_generation_params.pop("Hires resize", None)
return
if not state.processing_has_refined_job_count:
if state.job_count == -1:
state.job_count = self.n_iter
shared.total_tqdm.updateTotal((self.steps + (self.hr_second_pass_steps or self.steps)) * state.job_count)
state.job_count = state.job_count * 2
state.processing_has_refined_job_count = True
if self.hr_second_pass_steps:
self.extra_generation_params["Hires steps"] = self.hr_second_pass_steps
if self.hr_upscaler is not None:
self.extra_generation_params["Hires upscaler"] = self.hr_upscaler
def sample(self, conditioning, unconditional_conditioning, seeds, subseeds, subseed_strength, prompts):
self.sampler = sd_samplers.create_sampler(self.sampler_name, self.sd_model)
latent_scale_mode = shared.latent_upscale_modes.get(self.hr_upscaler, None) if self.hr_upscaler is not None else shared.latent_upscale_modes.get(shared.latent_upscale_default_mode, "nearest")
if self.enable_hr and latent_scale_mode is None:
assert len([x for x in shared.sd_upscalers if x.name == self.hr_upscaler]) > 0, f"could not find upscaler named {self.hr_upscaler}"
x = create_random_tensors([opt_C, self.height // opt_f, self.width // opt_f], seeds=seeds, subseeds=subseeds, subseed_strength=self.subseed_strength, seed_resize_from_h=self.seed_resize_from_h, seed_resize_from_w=self.seed_resize_from_w, p=self)
samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x))
if not self.enable_hr:
x = create_random_tensors([opt_C, self.height // opt_f, self.width // opt_f], seeds=seeds, subseeds=subseeds, subseed_strength=self.subseed_strength, seed_resize_from_h=self.seed_resize_from_h, seed_resize_from_w=self.seed_resize_from_w, p=self)
samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x))
return samples
x = create_random_tensors([opt_C, self.firstphase_height // opt_f, self.firstphase_width // opt_f], seeds=seeds, subseeds=subseeds, subseed_strength=self.subseed_strength, seed_resize_from_h=self.seed_resize_from_h, seed_resize_from_w=self.seed_resize_from_w, p=self)
samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x, self.firstphase_width, self.firstphase_height))
target_width = self.hr_upscale_to_x
target_height = self.hr_upscale_to_y
samples = samples[:, :, self.truncate_y//2:samples.shape[2]-self.truncate_y//2, self.truncate_x//2:samples.shape[3]-self.truncate_x//2]
"""saves image before applying hires fix, if enabled in options; takes as an argument either an image or batch with latent space images"""
def save_intermediate(image, index):
"""saves image before applying hires fix, if enabled in options; takes as an argument either an image or batch with latent space images"""
if not opts.save or self.do_not_save_samples or not opts.save_images_before_highres_fix:
return
if not isinstance(image, Image.Image):
image = sd_samplers.sample_to_image(image, index)
image = sd_samplers.sample_to_image(image, index, approximation=0)
images.save_image(image, self.outpath_samples, "", seeds[index], prompts[index], opts.samples_format, suffix="-before-highres-fix")
info = create_infotext(self, self.all_prompts, self.all_seeds, self.all_subseeds, [], iteration=self.iteration, position_in_batch=index)
images.save_image(image, self.outpath_samples, "", seeds[index], prompts[index], opts.samples_format, info=info, suffix="-before-highres-fix")
if opts.use_scale_latent_for_hires_fix:
if latent_scale_mode is not None:
for i in range(samples.shape[0]):
save_intermediate(samples, i)
samples = torch.nn.functional.interpolate(samples, size=(self.height // opt_f, self.width // opt_f), mode="bilinear")
samples = torch.nn.functional.interpolate(samples, size=(target_height // opt_f, target_width // opt_f), mode=latent_scale_mode["mode"], antialias=latent_scale_mode["antialias"])
# Avoid making the inpainting conditioning unless necessary as
# this does need some extra compute to decode / encode the image again.
@ -747,7 +838,7 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
save_intermediate(image, i)
image = images.resize_image(0, image, self.width, self.height)
image = images.resize_image(0, image, target_width, target_height, upscaler_name=self.hr_upscaler)
image = np.array(image).astype(np.float32) / 255.0
image = np.moveaxis(image, 2, 0)
batch_images.append(image)
@ -764,13 +855,15 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
self.sampler = sd_samplers.create_sampler(self.sampler_name, self.sd_model)
noise = create_random_tensors(samples.shape[1:], seeds=seeds, subseeds=subseeds, subseed_strength=subseed_strength, seed_resize_from_h=self.seed_resize_from_h, seed_resize_from_w=self.seed_resize_from_w, p=self)
samples = samples[:, :, self.truncate_y//2:samples.shape[2]-(self.truncate_y+1)//2, self.truncate_x//2:samples.shape[3]-(self.truncate_x+1)//2]
noise = create_random_tensors(samples.shape[1:], seeds=seeds, subseeds=subseeds, subseed_strength=subseed_strength, p=self)
# GC now before running the next img2img to prevent running out of memory
x = None
devices.torch_gc()
samples = self.sampler.sample_img2img(self, samples, noise, conditioning, unconditional_conditioning, steps=self.steps, image_conditioning=image_conditioning)
samples = self.sampler.sample_img2img(self, samples, noise, conditioning, unconditional_conditioning, steps=self.hr_second_pass_steps or self.steps, image_conditioning=image_conditioning)
return samples

18
modules/script_callbacks.py
View File

@ -71,6 +71,7 @@ callback_map = dict(
callbacks_before_component=[],
callbacks_after_component=[],
callbacks_image_grid=[],
callbacks_script_unloaded=[],
)
@ -171,6 +172,14 @@ def image_grid_callback(params: ImageGridLoopParams):
report_exception(c, 'image_grid')
def script_unloaded_callback():
for c in reversed(callback_map['callbacks_script_unloaded']):
try:
c.callback()
except Exception:
report_exception(c, 'script_unloaded')
def add_callback(callbacks, fun):
stack = [x for x in inspect.stack() if x.filename != __file__]
filename = stack[0].filename if len(stack) > 0 else 'unknown file'
@ -202,7 +211,7 @@ def on_app_started(callback):
def on_model_loaded(callback):
"""register a function to be called when the stable diffusion model is created; the model is
passed as an argument"""
passed as an argument; this function is also called when the script is reloaded. """
add_callback(callback_map['callbacks_model_loaded'], callback)
@ -279,3 +288,10 @@ def on_image_grid(callback):
- params: ImageGridLoopParams - parameters to be used for grid creation. Can be modified.
"""
add_callback(callback_map['callbacks_image_grid'], callback)
def on_script_unloaded(callback):
"""register a function to be called before the script is unloaded. Any hooks/hijacks/monkeying about that
the script did should be reverted here"""
add_callback(callback_map['callbacks_script_unloaded'], callback)

11
modules/scripts.py
View File

@ -1,4 +1,5 @@
import os
import re
import sys
import traceback
from collections import namedtuple
@ -128,6 +129,15 @@ class Script:
"""unused"""
return ""
def elem_id(self, item_id):
"""helper function to generate id for a HTML element, constructs final id out of script name, tab and user-supplied item_id"""
need_tabname = self.show(True) == self.show(False)
tabname = ('img2img' if self.is_img2img else 'txt2txt') + "_" if need_tabname else ""
title = re.sub(r'[^a-z_0-9]', '', re.sub(r'\s', '_', self.title().lower()))
return f'script_{tabname}{title}_{item_id}'
current_basedir = paths.script_path
@ -280,7 +290,6 @@ class ScriptRunner:
script.group = group
dropdown = gr.Dropdown(label="Script", elem_id="script_list", choices=["None"] + self.titles, value="None", type="index")
dropdown.save_to_config = True
inputs[0] = dropdown
for script in self.selectable_scripts:

95
modules/sd_disable_initialization.py Normal file
View File

@ -0,0 +1,95 @@
import ldm.modules.encoders.modules
import open_clip
import torch
import transformers.utils.hub
class DisableInitialization:
"""
When an object of this class enters a `with` block, it starts:
- preventing torch's layer initialization functions from working
- changes CLIP and OpenCLIP to not download model weights
- changes CLIP to not make requests to check if there is a new version of a file you already have
When it leaves the block, it reverts everything to how it was before.
Use it like this:
```
with DisableInitialization():
do_things()
```
"""
def __enter__(self):
def do_nothing(*args, **kwargs):
pass
def create_model_and_transforms_without_pretrained(*args, pretrained=None, **kwargs):
return self.create_model_and_transforms(*args, pretrained=None, **kwargs)
def CLIPTextModel_from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs):
return self.CLIPTextModel_from_pretrained(None, *model_args, config=pretrained_model_name_or_path, state_dict={}, **kwargs)
def transformers_modeling_utils_load_pretrained_model(*args, **kwargs):
args = args[0:3] + ('/', ) + args[4:] # resolved_archive_file; must set it to something to prevent what seems to be a bug
return self.transformers_modeling_utils_load_pretrained_model(*args, **kwargs)
def transformers_utils_hub_get_file_from_cache(original, url, *args, **kwargs):
# this file is always 404, prevent making request
if url == 'https://huggingface.co/openai/clip-vit-large-patch14/resolve/main/added_tokens.json':
raise transformers.utils.hub.EntryNotFoundError
try:
return original(url, *args, local_files_only=True, **kwargs)
except Exception as e:
return original(url, *args, local_files_only=False, **kwargs)
def transformers_utils_hub_get_from_cache(url, *args, local_files_only=False, **kwargs):
return transformers_utils_hub_get_file_from_cache(self.transformers_utils_hub_get_from_cache, url, *args, **kwargs)
def transformers_tokenization_utils_base_cached_file(url, *args, local_files_only=False, **kwargs):
return transformers_utils_hub_get_file_from_cache(self.transformers_tokenization_utils_base_cached_file, url, *args, **kwargs)
def transformers_configuration_utils_cached_file(url, *args, local_files_only=False, **kwargs):
return transformers_utils_hub_get_file_from_cache(self.transformers_configuration_utils_cached_file, url, *args, **kwargs)
self.init_kaiming_uniform = torch.nn.init.kaiming_uniform_
self.init_no_grad_normal = torch.nn.init._no_grad_normal_
self.init_no_grad_uniform_ = torch.nn.init._no_grad_uniform_
self.create_model_and_transforms = open_clip.create_model_and_transforms
self.CLIPTextModel_from_pretrained = ldm.modules.encoders.modules.CLIPTextModel.from_pretrained
self.transformers_modeling_utils_load_pretrained_model = getattr(transformers.modeling_utils.PreTrainedModel, '_load_pretrained_model', None)
self.transformers_tokenization_utils_base_cached_file = getattr(transformers.tokenization_utils_base, 'cached_file', None)
self.transformers_configuration_utils_cached_file = getattr(transformers.configuration_utils, 'cached_file', None)
self.transformers_utils_hub_get_from_cache = getattr(transformers.utils.hub, 'get_from_cache', None)
torch.nn.init.kaiming_uniform_ = do_nothing
torch.nn.init._no_grad_normal_ = do_nothing
torch.nn.init._no_grad_uniform_ = do_nothing
open_clip.create_model_and_transforms = create_model_and_transforms_without_pretrained
ldm.modules.encoders.modules.CLIPTextModel.from_pretrained = CLIPTextModel_from_pretrained
if self.transformers_modeling_utils_load_pretrained_model is not None:
transformers.modeling_utils.PreTrainedModel._load_pretrained_model = transformers_modeling_utils_load_pretrained_model
if self.transformers_tokenization_utils_base_cached_file is not None:
transformers.tokenization_utils_base.cached_file = transformers_tokenization_utils_base_cached_file
if self.transformers_configuration_utils_cached_file is not None:
transformers.configuration_utils.cached_file = transformers_configuration_utils_cached_file
if self.transformers_utils_hub_get_from_cache is not None:
transformers.utils.hub.get_from_cache = transformers_utils_hub_get_from_cache
def __exit__(self, exc_type, exc_val, exc_tb):
torch.nn.init.kaiming_uniform_ = self.init_kaiming_uniform
torch.nn.init._no_grad_normal_ = self.init_no_grad_normal
torch.nn.init._no_grad_uniform_ = self.init_no_grad_uniform_
open_clip.create_model_and_transforms = self.create_model_and_transforms
ldm.modules.encoders.modules.CLIPTextModel.from_pretrained = self.CLIPTextModel_from_pretrained
if self.transformers_modeling_utils_load_pretrained_model is not None:
transformers.modeling_utils.PreTrainedModel._load_pretrained_model = self.transformers_modeling_utils_load_pretrained_model
if self.transformers_tokenization_utils_base_cached_file is not None:
transformers.utils.hub.cached_file = self.transformers_tokenization_utils_base_cached_file
if self.transformers_configuration_utils_cached_file is not None:
transformers.utils.hub.cached_file = self.transformers_configuration_utils_cached_file
if self.transformers_utils_hub_get_from_cache is not None:
transformers.utils.hub.get_from_cache = self.transformers_utils_hub_get_from_cache

42
modules/sd_hijack.py
View File

@ -7,8 +7,6 @@ from modules.hypernetworks import hypernetwork
from modules.shared import cmd_opts
from modules import sd_hijack_clip, sd_hijack_open_clip, sd_hijack_unet, sd_hijack_xlmr, xlmr
from modules.sd_hijack_optimizations import invokeAI_mps_available
import ldm.modules.attention
import ldm.modules.diffusionmodules.model
import ldm.modules.diffusionmodules.openaimodel
@ -35,26 +33,34 @@ def apply_optimizations():
ldm.modules.diffusionmodules.model.nonlinearity = silu
ldm.modules.diffusionmodules.openaimodel.th = sd_hijack_unet.th
optimization_method = None
if cmd_opts.force_enable_xformers or (cmd_opts.xformers and shared.xformers_available and torch.version.cuda and (6, 0) <= torch.cuda.get_device_capability(shared.device) <= (9, 0)):
print("Applying xformers cross attention optimization.")
ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.xformers_attention_forward
ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.xformers_attnblock_forward
optimization_method = 'xformers'
elif cmd_opts.opt_sub_quad_attention:
print("Applying sub-quadratic cross attention optimization.")
ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.sub_quad_attention_forward
ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.sub_quad_attnblock_forward
optimization_method = 'sub-quadratic'
elif cmd_opts.opt_split_attention_v1:
print("Applying v1 cross attention optimization.")
ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.split_cross_attention_forward_v1
elif not cmd_opts.disable_opt_split_attention and (cmd_opts.opt_split_attention_invokeai or not torch.cuda.is_available()):
if not invokeAI_mps_available and shared.device.type == 'mps':
print("The InvokeAI cross attention optimization for MPS requires the psutil package which is not installed.")
print("Applying v1 cross attention optimization.")
ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.split_cross_attention_forward_v1
else:
print("Applying cross attention optimization (InvokeAI).")
ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.split_cross_attention_forward_invokeAI
optimization_method = 'V1'
elif not cmd_opts.disable_opt_split_attention and (cmd_opts.opt_split_attention_invokeai or not cmd_opts.opt_split_attention and not torch.cuda.is_available()):
print("Applying cross attention optimization (InvokeAI).")
ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.split_cross_attention_forward_invokeAI
optimization_method = 'InvokeAI'
elif not cmd_opts.disable_opt_split_attention and (cmd_opts.opt_split_attention or torch.cuda.is_available()):
print("Applying cross attention optimization (Doggettx).")
ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.split_cross_attention_forward
ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.cross_attention_attnblock_forward
optimization_method = 'Doggettx'
return optimization_method
def undo_optimizations():
@ -75,11 +81,14 @@ class StableDiffusionModelHijack:
layers = None
circular_enabled = False
clip = None
optimization_method = None
embedding_db = modules.textual_inversion.textual_inversion.EmbeddingDatabase(cmd_opts.embeddings_dir)
embedding_db = modules.textual_inversion.textual_inversion.EmbeddingDatabase()
def __init__(self):
self.embedding_db.add_embedding_dir(cmd_opts.embeddings_dir)
def hijack(self, m):
if type(m.cond_stage_model) == xlmr.BertSeriesModelWithTransformation:
model_embeddings = m.cond_stage_model.roberta.embeddings
model_embeddings.token_embedding = EmbeddingsWithFixes(model_embeddings.word_embeddings, self)
@ -94,7 +103,7 @@ class StableDiffusionModelHijack:
m.cond_stage_model.model.token_embedding = EmbeddingsWithFixes(m.cond_stage_model.model.token_embedding, self)
m.cond_stage_model = sd_hijack_open_clip.FrozenOpenCLIPEmbedderWithCustomWords(m.cond_stage_model, self)
apply_optimizations()
self.optimization_method = apply_optimizations()
self.clip = m.cond_stage_model
@ -110,7 +119,6 @@ class StableDiffusionModelHijack:
self.layers = flatten(m)
def undo_hijack(self, m):
if type(m.cond_stage_model) == xlmr.BertSeriesModelWithTransformation:
m.cond_stage_model = m.cond_stage_model.wrapped
@ -140,10 +148,10 @@ class StableDiffusionModelHijack:
def clear_comments(self):
self.comments = []
def tokenize(self, text):
_, remade_batch_tokens, _, _, _, token_count = self.clip.process_text([text])
def get_prompt_lengths(self, text):
_, token_count = self.clip.process_texts([text])
return remade_batch_tokens[0], token_count, sd_hijack_clip.get_target_prompt_token_count(token_count)
return token_count, self.clip.get_target_prompt_token_count(token_count)
class EmbeddingsWithFixes(torch.nn.Module):

347
modules/sd_hijack_clip.py
View File

@ -1,30 +1,89 @@
import math
from collections import namedtuple
import torch
from modules import prompt_parser, devices
from modules import prompt_parser, devices, sd_hijack
from modules.shared import opts
def get_target_prompt_token_count(token_count):
return math.ceil(max(token_count, 1) / 75) * 75
class PromptChunk:
"""
This object contains token ids, weight (multipliers:1.4) and textual inversion embedding info for a chunk of prompt.
If a prompt is short, it is represented by one PromptChunk, otherwise, multiple are necessary.
Each PromptChunk contains an exact amount of tokens - 77, which includes one for start and end token,
so just 75 tokens from prompt.
"""
def __init__(self):
self.tokens = []
self.multipliers = []
self.fixes = []
PromptChunkFix = namedtuple('PromptChunkFix', ['offset', 'embedding'])
"""An object of this type is a marker showing that textual inversion embedding's vectors have to placed at offset in the prompt
chunk. Thos objects are found in PromptChunk.fixes and, are placed into FrozenCLIPEmbedderWithCustomWordsBase.hijack.fixes, and finally
are applied by sd_hijack.EmbeddingsWithFixes's forward function."""
class FrozenCLIPEmbedderWithCustomWordsBase(torch.nn.Module):
"""A pytorch module that is a wrapper for FrozenCLIPEmbedder module. it enhances FrozenCLIPEmbedder, making it possible to
have unlimited prompt length and assign weights to tokens in prompt.
"""
def __init__(self, wrapped, hijack):
super().__init__()
self.wrapped = wrapped
self.hijack = hijack
"""Original FrozenCLIPEmbedder module; can also be FrozenOpenCLIPEmbedder or xlmr.BertSeriesModelWithTransformation,
depending on model."""
self.hijack: sd_hijack.StableDiffusionModelHijack = hijack
self.chunk_length = 75
def empty_chunk(self):
"""creates an empty PromptChunk and returns it"""
chunk = PromptChunk()
chunk.tokens = [self.id_start] + [self.id_end] * (self.chunk_length + 1)
chunk.multipliers = [1.0] * (self.chunk_length + 2)
return chunk
def get_target_prompt_token_count(self, token_count):
"""returns the maximum number of tokens a prompt of a known length can have before it requires one more PromptChunk to be represented"""
return math.ceil(max(token_count, 1) / self.chunk_length) * self.chunk_length
def tokenize(self, texts):
"""Converts a batch of texts into a batch of token ids"""
raise NotImplementedError
def encode_with_transformers(self, tokens):
"""
converts a batch of token ids (in python lists) into a single tensor with numeric respresentation of those tokens;
All python lists with tokens are assumed to have same length, usually 77.
if input is a list with B elements and each element has T tokens, expected output shape is (B, T, C), where C depends on
model - can be 768 and 1024.
Among other things, this call will read self.hijack.fixes, apply it to its inputs, and clear it (setting it to None).
"""
raise NotImplementedError
def encode_embedding_init_text(self, init_text, nvpt):
"""Converts text into a tensor with this text's tokens' embeddings. Note that those are embeddings before they are passed through
transformers. nvpt is used as a maximum length in tokens. If text produces less teokens than nvpt, only this many is returned."""
raise NotImplementedError
def tokenize_line(self, line, used_custom_terms, hijack_comments):
def tokenize_line(self, line):
"""
this transforms a single prompt into a list of PromptChunk objects - as many as needed to
represent the prompt.
Returns the list and the total number of tokens in the prompt.
"""
if opts.enable_emphasis:
parsed = prompt_parser.parse_prompt_attention(line)
else:
@ -32,205 +91,152 @@ class FrozenCLIPEmbedderWithCustomWordsBase(torch.nn.Module):
tokenized = self.tokenize([text for text, _ in parsed])
fixes = []
remade_tokens = []
multipliers = []
chunks = []
chunk = PromptChunk()
token_count = 0
last_comma = -1
for tokens, (text, weight) in zip(tokenized, parsed):
i = 0
while i < len(tokens):
token = tokens[i]
def next_chunk():
"""puts current chunk into the list of results and produces the next one - empty"""
nonlocal token_count
nonlocal last_comma
nonlocal chunk
embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i)
token_count += len(chunk.tokens)
to_add = self.chunk_length - len(chunk.tokens)
if to_add > 0:
chunk.tokens += [self.id_end] * to_add
chunk.multipliers += [1.0] * to_add
chunk.tokens = [self.id_start] + chunk.tokens + [self.id_end]
chunk.multipliers = [1.0] + chunk.multipliers + [1.0]
last_comma = -1
chunks.append(chunk)
chunk = PromptChunk()
for tokens, (text, weight) in zip(tokenized, parsed):
position = 0
while position < len(tokens):
token = tokens[position]
if token == self.comma_token:
last_comma = len(remade_tokens)
elif opts.comma_padding_backtrack != 0 and max(len(remade_tokens), 1) % 75 == 0 and last_comma != -1 and len(remade_tokens) - last_comma <= opts.comma_padding_backtrack:
last_comma += 1
reloc_tokens = remade_tokens[last_comma:]
reloc_mults = multipliers[last_comma:]
last_comma = len(chunk.tokens)
remade_tokens = remade_tokens[:last_comma]
length = len(remade_tokens)
# this is when we are at the end of alloted 75 tokens for the current chunk, and the current token is not a comma. opts.comma_padding_backtrack
# is a setting that specifies that if there is a comma nearby, the text after the comma should be moved out of this chunk and into the next.
elif opts.comma_padding_backtrack != 0 and len(chunk.tokens) == self.chunk_length and last_comma != -1 and len(chunk.tokens) - last_comma <= opts.comma_padding_backtrack:
break_location = last_comma + 1
rem = int(math.ceil(length / 75)) * 75 - length
remade_tokens += [self.id_end] * rem + reloc_tokens
multipliers = multipliers[:last_comma] + [1.0] * rem + reloc_mults
reloc_tokens = chunk.tokens[break_location:]
reloc_mults = chunk.multipliers[break_location:]
chunk.tokens = chunk.tokens[:break_location]
chunk.multipliers = chunk.multipliers[:break_location]
next_chunk()
chunk.tokens = reloc_tokens
chunk.multipliers = reloc_mults
if len(chunk.tokens) == self.chunk_length:
next_chunk()
embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, position)
if embedding is None:
remade_tokens.append(token)
multipliers.append(weight)
i += 1
else:
emb_len = int(embedding.vec.shape[0])
iteration = len(remade_tokens) // 75
if (len(remade_tokens) + emb_len) // 75 != iteration:
rem = (75 * (iteration + 1) - len(remade_tokens))
remade_tokens += [self.id_end] * rem
multipliers += [1.0] * rem
iteration += 1
fixes.append((iteration, (len(remade_tokens) % 75, embedding)))
remade_tokens += [0] * emb_len
multipliers += [weight] * emb_len
used_custom_terms.append((embedding.name, embedding.checksum()))
i += embedding_length_in_tokens
chunk.tokens.append(token)
chunk.multipliers.append(weight)
position += 1
continue
token_count = len(remade_tokens)
prompt_target_length = get_target_prompt_token_count(token_count)
tokens_to_add = prompt_target_length - len(remade_tokens)
emb_len = int(embedding.vec.shape[0])
if len(chunk.tokens) + emb_len > self.chunk_length:
next_chunk()
remade_tokens = remade_tokens + [self.id_end] * tokens_to_add
multipliers = multipliers + [1.0] * tokens_to_add
chunk.fixes.append(PromptChunkFix(len(chunk.tokens), embedding))
return remade_tokens, fixes, multipliers, token_count
chunk.tokens += [0] * emb_len
chunk.multipliers += [weight] * emb_len
position += embedding_length_in_tokens
if len(chunk.tokens) > 0 or len(chunks) == 0:
next_chunk()
return chunks, token_count
def process_texts(self, texts):
"""
Accepts a list of texts and calls tokenize_line() on each, with cache. Returns the list of results and maximum
length, in tokens, of all texts.
"""
def process_text(self, texts):
used_custom_terms = []
remade_batch_tokens = []
hijack_comments = []
hijack_fixes = []
token_count = 0
cache = {}
batch_multipliers = []
batch_chunks = []
for line in texts:
if line in cache:
remade_tokens, fixes, multipliers = cache[line]
chunks = cache[line]
else:
remade_tokens, fixes, multipliers, current_token_count = self.tokenize_line(line, used_custom_terms, hijack_comments)
chunks, current_token_count = self.tokenize_line(line)
token_count = max(current_token_count, token_count)
cache[line] = (remade_tokens, fixes, multipliers)
cache[line] = chunks
remade_batch_tokens.append(remade_tokens)
hijack_fixes.append(fixes)
batch_multipliers.append(multipliers)
batch_chunks.append(chunks)
return batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count
return batch_chunks, token_count
def process_text_old(self, texts):
id_start = self.id_start
id_end = self.id_end
maxlen = self.wrapped.max_length # you get to stay at 77
used_custom_terms = []
remade_batch_tokens = []
hijack_comments = []
hijack_fixes = []
token_count = 0
def forward(self, texts):
"""
Accepts an array of texts; Passes texts through transformers network to create a tensor with numerical representation of those texts.
Returns a tensor with shape of (B, T, C), where B is length of the array; T is length, in tokens, of texts (including padding) - T will
be a multiple of 77; and C is dimensionality of each token - for SD1 it's 768, and for SD2 it's 1024.
An example shape returned by this function can be: (2, 77, 768).
Webui usually sends just one text at a time through this function - the only time when texts is an array with more than one elemenet
is when you do prompt editing: "a picture of a [cat:dog:0.4] eating ice cream"
"""
cache = {}
batch_tokens = self.tokenize(texts)
batch_multipliers = []
for tokens in batch_tokens:
tuple_tokens = tuple(tokens)
if opts.use_old_emphasis_implementation:
import modules.sd_hijack_clip_old
return modules.sd_hijack_clip_old.forward_old(self, texts)
if tuple_tokens in cache:
remade_tokens, fixes, multipliers = cache[tuple_tokens]
else:
fixes = []
remade_tokens = []
multipliers = []
mult = 1.0
batch_chunks, token_count = self.process_texts(texts)
i = 0
while i < len(tokens):
token = tokens[i]
used_embeddings = {}
chunk_count = max([len(x) for x in batch_chunks])
embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i)
zs = []
for i in range(chunk_count):
batch_chunk = [chunks[i] if i < len(chunks) else self.empty_chunk() for chunks in batch_chunks]
mult_change = self.token_mults.get(token) if opts.enable_emphasis else None
if mult_change is not None:
mult *= mult_change
i += 1
elif embedding is None:
remade_tokens.append(token)
multipliers.append(mult)
i += 1
else:
emb_len = int(embedding.vec.shape[0])
fixes.append((len(remade_tokens), embedding))
remade_tokens += [0] * emb_len
multipliers += [mult] * emb_len
used_custom_terms.append((embedding.name, embedding.checksum()))
i += embedding_length_in_tokens
tokens = [x.tokens for x in batch_chunk]
multipliers = [x.multipliers for x in batch_chunk]
self.hijack.fixes = [x.fixes for x in batch_chunk]
if len(remade_tokens) > maxlen - 2:
vocab = {v: k for k, v in self.wrapped.tokenizer.get_vocab().items()}
ovf = remade_tokens[maxlen - 2:]
overflowing_words = [vocab.get(int(x), "") for x in ovf]
overflowing_text = self.wrapped.tokenizer.convert_tokens_to_string(''.join(overflowing_words))
hijack_comments.append(f"Warning: too many input tokens; some ({len(overflowing_words)}) have been truncated:\n{overflowing_text}\n")
for fixes in self.hijack.fixes:
for position, embedding in fixes:
used_embeddings[embedding.name] = embedding
token_count = len(remade_tokens)
remade_tokens = remade_tokens + [id_end] * (maxlen - 2 - len(remade_tokens))
remade_tokens = [id_start] + remade_tokens[0:maxlen - 2] + [id_end]
cache[tuple_tokens] = (remade_tokens, fixes, multipliers)
z = self.process_tokens(tokens, multipliers)
zs.append(z)
multipliers = multipliers + [1.0] * (maxlen - 2 - len(multipliers))
multipliers = [1.0] + multipliers[0:maxlen - 2] + [1.0]
if len(used_embeddings) > 0:
embeddings_list = ", ".join([f'{name} [{embedding.checksum()}]' for name, embedding in used_embeddings.items()])
self.hijack.comments.append(f"Used embeddings: {embeddings_list}")
remade_batch_tokens.append(remade_tokens)
hijack_fixes.append(fixes)
batch_multipliers.append(multipliers)
return batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count
def forward(self, text):
use_old = opts.use_old_emphasis_implementation
if use_old:
batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = self.process_text_old(text)
else:
batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = self.process_text(text)
self.hijack.comments += hijack_comments
if len(used_custom_terms) > 0:
self.hijack.comments.append("Used embeddings: " + ", ".join([f'{word} [{checksum}]' for word, checksum in used_custom_terms]))
if use_old:
self.hijack.fixes = hijack_fixes
return self.process_tokens(remade_batch_tokens, batch_multipliers)
z = None
i = 0
while max(map(len, remade_batch_tokens)) != 0:
rem_tokens = [x[75:] for x in remade_batch_tokens]
rem_multipliers = [x[75:] for x in batch_multipliers]
self.hijack.fixes = []
for unfiltered in hijack_fixes:
fixes = []
for fix in unfiltered:
if fix[0] == i:
fixes.append(fix[1])
self.hijack.fixes.append(fixes)
tokens = []
multipliers = []
for j in range(len(remade_batch_tokens)):
if len(remade_batch_tokens[j]) > 0:
tokens.append(remade_batch_tokens[j][:75])
multipliers.append(batch_multipliers[j][:75])
else:
tokens.append([self.id_end] * 75)
multipliers.append([1.0] * 75)
z1 = self.process_tokens(tokens, multipliers)
z = z1 if z is None else torch.cat((z, z1), axis=-2)
remade_batch_tokens = rem_tokens
batch_multipliers = rem_multipliers
i += 1
return z
return torch.hstack(zs)
def process_tokens(self, remade_batch_tokens, batch_multipliers):
if not opts.use_old_emphasis_implementation:
remade_batch_tokens = [[self.id_start] + x[:75] + [self.id_end] for x in remade_batch_tokens]
batch_multipliers = [[1.0] + x[:75] + [1.0] for x in batch_multipliers]
"""
sends one single prompt chunk to be encoded by transformers neural network.
remade_batch_tokens is a batch of tokens - a list, where every element is a list of tokens; usually
there are exactly 77 tokens in the list. batch_multipliers is the same but for multipliers instead of tokens.
Multipliers are used to give more or less weight to the outputs of transformers network. Each multiplier
corresponds to one token.
"""
tokens = torch.asarray(remade_batch_tokens).to(devices.device)
# this is for SD2: SD1 uses the same token for padding and end of text, while SD2 uses different ones.
if self.id_end != self.id_pad:
for batch_pos in range(len(remade_batch_tokens)):
index = remade_batch_tokens[batch_pos].index(self.id_end)
@ -239,12 +245,11 @@ class FrozenCLIPEmbedderWithCustomWordsBase(torch.nn.Module):
z = self.encode_with_transformers(tokens)
# restoring original mean is likely not correct, but it seems to work well to prevent artifacts that happen otherwise
batch_multipliers_of_same_length = [x + [1.0] * (75 - len(x)) for x in batch_multipliers]
batch_multipliers = torch.asarray(batch_multipliers_of_same_length).to(devices.device)
batch_multipliers = torch.asarray(batch_multipliers).to(devices.device)
original_mean = z.mean()
z *= batch_multipliers.reshape(batch_multipliers.shape + (1,)).expand(z.shape)
z = z * batch_multipliers.reshape(batch_multipliers.shape + (1,)).expand(z.shape)
new_mean = z.mean()
z *= original_mean / new_mean
z = z * (original_mean / new_mean)
return z

81
modules/sd_hijack_clip_old.py Normal file
View File

@ -0,0 +1,81 @@
from modules import sd_hijack_clip
from modules import shared
def process_text_old(self: sd_hijack_clip.FrozenCLIPEmbedderWithCustomWordsBase, texts):
id_start = self.id_start
id_end = self.id_end
maxlen = self.wrapped.max_length # you get to stay at 77
used_custom_terms = []
remade_batch_tokens = []
hijack_comments = []
hijack_fixes = []
token_count = 0
cache = {}
batch_tokens = self.tokenize(texts)
batch_multipliers = []
for tokens in batch_tokens:
tuple_tokens = tuple(tokens)
if tuple_tokens in cache:
remade_tokens, fixes, multipliers = cache[tuple_tokens]
else:
fixes = []
remade_tokens = []
multipliers = []
mult = 1.0
i = 0
while i < len(tokens):
token = tokens[i]
embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i)
mult_change = self.token_mults.get(token) if shared.opts.enable_emphasis else None
if mult_change is not None:
mult *= mult_change
i += 1
elif embedding is None:
remade_tokens.append(token)
multipliers.append(mult)
i += 1
else:
emb_len = int(embedding.vec.shape[0])
fixes.append((len(remade_tokens), embedding))
remade_tokens += [0] * emb_len
multipliers += [mult] * emb_len
used_custom_terms.append((embedding.name, embedding.checksum()))
i += embedding_length_in_tokens
if len(remade_tokens) > maxlen - 2:
vocab = {v: k for k, v in self.wrapped.tokenizer.get_vocab().items()}
ovf = remade_tokens[maxlen - 2:]
overflowing_words = [vocab.get(int(x), "") for x in ovf]
overflowing_text = self.wrapped.tokenizer.convert_tokens_to_string(''.join(overflowing_words))
hijack_comments.append(f"Warning: too many input tokens; some ({len(overflowing_words)}) have been truncated:\n{overflowing_text}\n")
token_count = len(remade_tokens)
remade_tokens = remade_tokens + [id_end] * (maxlen - 2 - len(remade_tokens))
remade_tokens = [id_start] + remade_tokens[0:maxlen - 2] + [id_end]
cache[tuple_tokens] = (remade_tokens, fixes, multipliers)
multipliers = multipliers + [1.0] * (maxlen - 2 - len(multipliers))
multipliers = [1.0] + multipliers[0:maxlen - 2] + [1.0]
remade_batch_tokens.append(remade_tokens)
hijack_fixes.append(fixes)
batch_multipliers.append(multipliers)
return batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count
def forward_old(self: sd_hijack_clip.FrozenCLIPEmbedderWithCustomWordsBase, texts):
batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = process_text_old(self, texts)
self.hijack.comments += hijack_comments
if len(used_custom_terms) > 0:
self.hijack.comments.append("Used embeddings: " + ", ".join([f'{word} [{checksum}]' for word, checksum in used_custom_terms]))
self.hijack.fixes = hijack_fixes
return self.process_tokens(remade_batch_tokens, batch_multipliers)

237
modules/sd_hijack_inpainting.py
View File

@ -12,191 +12,6 @@ from ldm.models.diffusion.ddpm import LatentDiffusion
from ldm.models.diffusion.plms import PLMSSampler
from ldm.models.diffusion.ddim import DDIMSampler, noise_like
# =================================================================================================
# Monkey patch DDIMSampler methods from RunwayML repo directly.
# Adapted from:
# https://github.com/runwayml/stable-diffusion/blob/main/ldm/models/diffusion/ddim.py
# =================================================================================================
@torch.no_grad()
def sample_ddim(self,
S,
batch_size,
shape,
conditioning=None,
callback=None,
normals_sequence=None,
img_callback=None,
quantize_x0=False,
eta=0.,
mask=None,
x0=None,
temperature=1.,
noise_dropout=0.,
score_corrector=None,
corrector_kwargs=None,
verbose=True,
x_T=None,
log_every_t=100,
unconditional_guidance_scale=1.,
unconditional_conditioning=None,
# this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
**kwargs
):
if conditioning is not None:
if isinstance(conditioning, dict):
ctmp = conditioning[list(conditioning.keys())[0]]
while isinstance(ctmp, list):
ctmp = ctmp[0]
cbs = ctmp.shape[0]
if cbs != batch_size:
print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
else:
if conditioning.shape[0] != batch_size:
print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
# sampling
C, H, W = shape
size = (batch_size, C, H, W)
print(f'Data shape for DDIM sampling is {size}, eta {eta}')
samples, intermediates = self.ddim_sampling(conditioning, size,
callback=callback,
img_callback=img_callback,
quantize_denoised=quantize_x0,
mask=mask, x0=x0,
ddim_use_original_steps=False,
noise_dropout=noise_dropout,
temperature=temperature,
score_corrector=score_corrector,
corrector_kwargs=corrector_kwargs,
x_T=x_T,
log_every_t=log_every_t,
unconditional_guidance_scale=unconditional_guidance_scale,
unconditional_conditioning=unconditional_conditioning,
)
return samples, intermediates
@torch.no_grad()
def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
unconditional_guidance_scale=1., unconditional_conditioning=None):
b, *_, device = *x.shape, x.device
if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
e_t = self.model.apply_model(x, t, c)
else:
x_in = torch.cat([x] * 2)
t_in = torch.cat([t] * 2)
if isinstance(c, dict):
assert isinstance(unconditional_conditioning, dict)
c_in = dict()
for k in c:
if isinstance(c[k], list):
c_in[k] = [
torch.cat([unconditional_conditioning[k][i], c[k][i]])
for i in range(len(c[k]))
]
else:
c_in[k] = torch.cat([unconditional_conditioning[k], c[k]])
else:
c_in = torch.cat([unconditional_conditioning, c])
e_t_uncond, e_t = self.model.apply_model(x_in, t_in, c_in).chunk(2)
e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
if score_corrector is not None:
assert self.model.parameterization == "eps"
e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
# select parameters corresponding to the currently considered timestep
a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device)
# current prediction for x_0
pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
if quantize_denoised:
pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
# direction pointing to x_t
dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
if noise_dropout > 0.:
noise = torch.nn.functional.dropout(noise, p=noise_dropout)
x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
return x_prev, pred_x0
# =================================================================================================
# Monkey patch PLMSSampler methods.
# This one was not actually patched correctly in the RunwayML repo, but we can replicate the changes.
# Adapted from:
# https://github.com/CompVis/stable-diffusion/blob/main/ldm/models/diffusion/plms.py
# =================================================================================================
@torch.no_grad()
def sample_plms(self,
S,
batch_size,
shape,
conditioning=None,
callback=None,
normals_sequence=None,
img_callback=None,
quantize_x0=False,
eta=0.,
mask=None,
x0=None,
temperature=1.,
noise_dropout=0.,
score_corrector=None,
corrector_kwargs=None,
verbose=True,
x_T=None,
log_every_t=100,
unconditional_guidance_scale=1.,
unconditional_conditioning=None,
# this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
**kwargs
):
if conditioning is not None:
if isinstance(conditioning, dict):
ctmp = conditioning[list(conditioning.keys())[0]]
while isinstance(ctmp, list):
ctmp = ctmp[0]
cbs = ctmp.shape[0]
if cbs != batch_size:
print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
else:
if conditioning.shape[0] != batch_size:
print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
# sampling
C, H, W = shape
size = (batch_size, C, H, W)
# print(f'Data shape for PLMS sampling is {size}') # remove unnecessary message
samples, intermediates = self.plms_sampling(conditioning, size,
callback=callback,
img_callback=img_callback,
quantize_denoised=quantize_x0,
mask=mask, x0=x0,
ddim_use_original_steps=False,
noise_dropout=noise_dropout,
temperature=temperature,
score_corrector=score_corrector,
corrector_kwargs=corrector_kwargs,
x_T=x_T,
log_every_t=log_every_t,
unconditional_guidance_scale=unconditional_guidance_scale,
unconditional_conditioning=unconditional_conditioning,
)
return samples, intermediates
@torch.no_grad()
def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
@ -280,61 +95,17 @@ def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=F
return x_prev, pred_x0, e_t
# =================================================================================================
# Monkey patch LatentInpaintDiffusion to load the checkpoint with a proper config.
# Adapted from:
# https://github.com/runwayml/stable-diffusion/blob/main/ldm/models/diffusion/ddpm.py
# =================================================================================================
@torch.no_grad()
def get_unconditional_conditioning(self, batch_size, null_label=None):
if null_label is not None:
xc = null_label
if isinstance(xc, ListConfig):
xc = list(xc)
if isinstance(xc, dict) or isinstance(xc, list):
c = self.get_learned_conditioning(xc)
else:
if hasattr(xc, "to"):
xc = xc.to(self.device)
c = self.get_learned_conditioning(xc)
else:
# todo: get null label from cond_stage_model
raise NotImplementedError()
c = repeat(c, "1 ... -> b ...", b=batch_size).to(self.device)
return c
class LatentInpaintDiffusion(LatentDiffusion):
def __init__(
self,
concat_keys=("mask", "masked_image"),
masked_image_key="masked_image",
*args,
**kwargs,
):
super().__init__(*args, **kwargs)
self.masked_image_key = masked_image_key
assert self.masked_image_key in concat_keys
self.concat_keys = concat_keys
def should_hijack_inpainting(checkpoint_info):
from modules import sd_models
ckpt_basename = os.path.basename(checkpoint_info.filename).lower()
cfg_basename = os.path.basename(checkpoint_info.config).lower()
cfg_basename = os.path.basename(sd_models.find_checkpoint_config(checkpoint_info)).lower()
return "inpainting" in ckpt_basename and not "inpainting" in cfg_basename
def do_inpainting_hijack():
# most of this stuff seems to no longer be needed because it is already included into SD2.0
# p_sample_plms is needed because PLMS can't work with dicts as conditionings
# this file should be cleaned up later if everything turns out to work fine
# ldm.models.diffusion.ddpm.get_unconditional_conditioning = get_unconditional_conditioning
# ldm.models.diffusion.ddpm.LatentInpaintDiffusion = LatentInpaintDiffusion
# ldm.models.diffusion.ddim.DDIMSampler.p_sample_ddim = p_sample_ddim
# ldm.models.diffusion.ddim.DDIMSampler.sample = sample_ddim
ldm.models.diffusion.plms.PLMSSampler.p_sample_plms = p_sample_plms
# ldm.models.diffusion.plms.PLMSSampler.sample = sample_plms

125
modules/sd_hijack_optimizations.py
View File

@ -1,7 +1,7 @@
import math
import sys
import traceback
import importlib
import psutil
import torch
from torch import einsum
@ -12,6 +12,8 @@ from einops import rearrange
from modules import shared
from modules.hypernetworks import hypernetwork
from .sub_quadratic_attention import efficient_dot_product_attention
if shared.cmd_opts.xformers or shared.cmd_opts.force_enable_xformers:
try:
@ -22,6 +24,19 @@ if shared.cmd_opts.xformers or shared.cmd_opts.force_enable_xformers:
print(traceback.format_exc(), file=sys.stderr)
def get_available_vram():
if shared.device.type == 'cuda':
stats = torch.cuda.memory_stats(shared.device)
mem_active = stats['active_bytes.all.current']
mem_reserved = stats['reserved_bytes.all.current']
mem_free_cuda, _ = torch.cuda.mem_get_info(torch.cuda.current_device())
mem_free_torch = mem_reserved - mem_active
mem_free_total = mem_free_cuda + mem_free_torch
return mem_free_total
else:
return psutil.virtual_memory().available
# see https://github.com/basujindal/stable-diffusion/pull/117 for discussion
def split_cross_attention_forward_v1(self, x, context=None, mask=None):
h = self.heads
@ -76,12 +91,7 @@ def split_cross_attention_forward(self, x, context=None, mask=None):
r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device, dtype=q.dtype)
stats = torch.cuda.memory_stats(q.device)
mem_active = stats['active_bytes.all.current']
mem_reserved = stats['reserved_bytes.all.current']
mem_free_cuda, _ = torch.cuda.mem_get_info(torch.cuda.current_device())
mem_free_torch = mem_reserved - mem_active
mem_free_total = mem_free_cuda + mem_free_torch
mem_free_total = get_available_vram()
gb = 1024 ** 3
tensor_size = q.shape[0] * q.shape[1] * k.shape[1] * q.element_size()
@ -118,19 +128,8 @@ def split_cross_attention_forward(self, x, context=None, mask=None):
return self.to_out(r2)
def check_for_psutil():
try:
spec = importlib.util.find_spec('psutil')
return spec is not None
except ModuleNotFoundError:
return False
invokeAI_mps_available = check_for_psutil()
# -- Taken from https://github.com/invoke-ai/InvokeAI and modified --
if invokeAI_mps_available:
import psutil
mem_total_gb = psutil.virtual_memory().total // (1 << 30)
mem_total_gb = psutil.virtual_memory().total // (1 << 30)
def einsum_op_compvis(q, k, v):
s = einsum('b i d, b j d -> b i j', q, k)
@ -215,6 +214,71 @@ def split_cross_attention_forward_invokeAI(self, x, context=None, mask=None):
# -- End of code from https://github.com/invoke-ai/InvokeAI --
# Based on Birch-san's modified implementation of sub-quadratic attention from https://github.com/Birch-san/diffusers/pull/1
# The sub_quad_attention_forward function is under the MIT License listed under Memory Efficient Attention in the Licenses section of the web UI interface
def sub_quad_attention_forward(self, x, context=None, mask=None):
assert mask is None, "attention-mask not currently implemented for SubQuadraticCrossAttnProcessor."
h = self.heads
q = self.to_q(x)
context = default(context, x)
context_k, context_v = hypernetwork.apply_hypernetwork(shared.loaded_hypernetwork, context)
k = self.to_k(context_k)
v = self.to_v(context_v)
del context, context_k, context_v, x
q = q.unflatten(-1, (h, -1)).transpose(1,2).flatten(end_dim=1)
k = k.unflatten(-1, (h, -1)).transpose(1,2).flatten(end_dim=1)
v = v.unflatten(-1, (h, -1)).transpose(1,2).flatten(end_dim=1)
x = sub_quad_attention(q, k, v, q_chunk_size=shared.cmd_opts.sub_quad_q_chunk_size, kv_chunk_size=shared.cmd_opts.sub_quad_kv_chunk_size, chunk_threshold=shared.cmd_opts.sub_quad_chunk_threshold, use_checkpoint=self.training)
x = x.unflatten(0, (-1, h)).transpose(1,2).flatten(start_dim=2)
out_proj, dropout = self.to_out
x = out_proj(x)
x = dropout(x)
return x
def sub_quad_attention(q, k, v, q_chunk_size=1024, kv_chunk_size=None, kv_chunk_size_min=None, chunk_threshold=None, use_checkpoint=True):
bytes_per_token = torch.finfo(q.dtype).bits//8
batch_x_heads, q_tokens, _ = q.shape
_, k_tokens, _ = k.shape
qk_matmul_size_bytes = batch_x_heads * bytes_per_token * q_tokens * k_tokens
if chunk_threshold is None:
chunk_threshold_bytes = int(get_available_vram() * 0.9) if q.device.type == 'mps' else int(get_available_vram() * 0.7)
elif chunk_threshold == 0:
chunk_threshold_bytes = None
else:
chunk_threshold_bytes = int(0.01 * chunk_threshold * get_available_vram())
if kv_chunk_size_min is None and chunk_threshold_bytes is not None:
kv_chunk_size_min = chunk_threshold_bytes // (batch_x_heads * bytes_per_token * (k.shape[2] + v.shape[2]))
elif kv_chunk_size_min == 0:
kv_chunk_size_min = None
if chunk_threshold_bytes is not None and qk_matmul_size_bytes <= chunk_threshold_bytes:
# the big matmul fits into our memory limit; do everything in 1 chunk,
# i.e. send it down the unchunked fast-path
query_chunk_size = q_tokens
kv_chunk_size = k_tokens
return efficient_dot_product_attention(
q,
k,
v,
query_chunk_size=q_chunk_size,
kv_chunk_size=kv_chunk_size,
kv_chunk_size_min = kv_chunk_size_min,
use_checkpoint=use_checkpoint,
)
def xformers_attention_forward(self, x, context=None, mask=None):
h = self.heads
q_in = self.to_q(x)
@ -252,12 +316,7 @@ def cross_attention_attnblock_forward(self, x):
h_ = torch.zeros_like(k, device=q.device)
stats = torch.cuda.memory_stats(q.device)
mem_active = stats['active_bytes.all.current']
mem_reserved = stats['reserved_bytes.all.current']
mem_free_cuda, _ = torch.cuda.mem_get_info(torch.cuda.current_device())
mem_free_torch = mem_reserved - mem_active
mem_free_total = mem_free_cuda + mem_free_torch
mem_free_total = get_available_vram()
tensor_size = q.shape[0] * q.shape[1] * k.shape[2] * q.element_size()
mem_required = tensor_size * 2.5
@ -312,3 +371,19 @@ def xformers_attnblock_forward(self, x):
return x + out
except NotImplementedError:
return cross_attention_attnblock_forward(self, x)
def sub_quad_attnblock_forward(self, x):
h_ = x
h_ = self.norm(h_)
q = self.q(h_)
k = self.k(h_)
v = self.v(h_)
b, c, h, w = q.shape
q, k, v = map(lambda t: rearrange(t, 'b c h w -> b (h w) c'), (q, k, v))
q = q.contiguous()
k = k.contiguous()
v = v.contiguous()
out = sub_quad_attention(q, k, v, q_chunk_size=shared.cmd_opts.sub_quad_q_chunk_size, kv_chunk_size=shared.cmd_opts.sub_quad_kv_chunk_size, chunk_threshold=shared.cmd_opts.sub_quad_chunk_threshold, use_checkpoint=self.training)
out = rearrange(out, 'b (h w) c -> b c h w', h=h)
out = self.proj_out(out)
return x + out

111
modules/sd_models.py
View File

@ -2,6 +2,7 @@ import collections
import os.path
import sys
import gc
import time
from collections import namedtuple
import torch
import re
@ -13,14 +14,14 @@ import ldm.modules.midas as midas
from ldm.util import instantiate_from_config
from modules import shared, modelloader, devices, script_callbacks, sd_vae
from modules import shared, modelloader, devices, script_callbacks, sd_vae, sd_disable_initialization, errors
from modules.paths import models_path
from modules.sd_hijack_inpainting import do_inpainting_hijack, should_hijack_inpainting
model_dir = "Stable-diffusion"
model_path = os.path.abspath(os.path.join(models_path, model_dir))
CheckpointInfo = namedtuple("CheckpointInfo", ['filename', 'title', 'hash', 'model_name', 'config'])
CheckpointInfo = namedtuple("CheckpointInfo", ['filename', 'title', 'hash', 'model_name'])
checkpoints_list = {}
checkpoints_loaded = collections.OrderedDict()
@ -48,9 +49,20 @@ def checkpoint_tiles():
return sorted([x.title for x in checkpoints_list.values()], key = alphanumeric_key)
def find_checkpoint_config(info):
if info is None:
return shared.cmd_opts.config
config = os.path.splitext(info.filename)[0] + ".yaml"
if os.path.exists(config):
return config
return shared.cmd_opts.config
def list_models():
checkpoints_list.clear()
model_list = modelloader.load_models(model_path=model_path, command_path=shared.cmd_opts.ckpt_dir, ext_filter=[".ckpt", ".safetensors"])
model_list = modelloader.load_models(model_path=model_path, command_path=shared.cmd_opts.ckpt_dir, ext_filter=[".ckpt", ".safetensors"], ext_blacklist=[".vae.safetensors"])
def modeltitle(path, shorthash):
abspath = os.path.abspath(path)
@ -73,7 +85,7 @@ def list_models():
if os.path.exists(cmd_ckpt):
h = model_hash(cmd_ckpt)
title, short_model_name = modeltitle(cmd_ckpt, h)
checkpoints_list[title] = CheckpointInfo(cmd_ckpt, title, h, short_model_name, shared.cmd_opts.config)
checkpoints_list[title] = CheckpointInfo(cmd_ckpt, title, h, short_model_name)
shared.opts.data['sd_model_checkpoint'] = title
elif cmd_ckpt is not None and cmd_ckpt != shared.default_sd_model_file:
print(f"Checkpoint in --ckpt argument not found (Possible it was moved to {model_path}: {cmd_ckpt}", file=sys.stderr)
@ -81,12 +93,7 @@ def list_models():
h = model_hash(filename)
title, short_model_name = modeltitle(filename, h)
basename, _ = os.path.splitext(filename)
config = basename + ".yaml"
if not os.path.exists(config):
config = shared.cmd_opts.config
checkpoints_list[title] = CheckpointInfo(filename, title, h, short_model_name, config)
checkpoints_list[title] = CheckpointInfo(filename, title, h, short_model_name)
def get_closet_checkpoint_match(searchString):
@ -168,7 +175,10 @@ def get_state_dict_from_checkpoint(pl_sd):
def read_state_dict(checkpoint_file, print_global_state=False, map_location=None):
_, extension = os.path.splitext(checkpoint_file)
if extension.lower() == ".safetensors":
pl_sd = safetensors.torch.load_file(checkpoint_file, device=map_location or shared.weight_load_location)
device = map_location or shared.weight_load_location
if device is None:
device = devices.get_cuda_device_string() if torch.cuda.is_available() else "cpu"
pl_sd = safetensors.torch.load_file(checkpoint_file, device=device)
else:
pl_sd = torch.load(checkpoint_file, map_location=map_location or shared.weight_load_location)
@ -228,6 +238,8 @@ def load_model_weights(model, checkpoint_info, vae_file="auto"):
model.sd_model_checkpoint = checkpoint_file
model.sd_checkpoint_info = checkpoint_info
model.logvar = model.logvar.to(devices.device) # fix for training
sd_vae.delete_base_vae()
sd_vae.clear_loaded_vae()
vae_file = sd_vae.resolve_vae(checkpoint_file, vae_file=vae_file)
@ -276,12 +288,25 @@ def enable_midas_autodownload():
midas.api.load_model = load_model_wrapper
class Timer:
def __init__(self):
self.start = time.time()
def elapsed(self):
end = time.time()
res = end - self.start
self.start = end
return res
def load_model(checkpoint_info=None):
from modules import lowvram, sd_hijack
checkpoint_info = checkpoint_info or select_checkpoint()
checkpoint_config = find_checkpoint_config(checkpoint_info)
if checkpoint_info.config != shared.cmd_opts.config:
print(f"Loading config from: {checkpoint_info.config}")
if checkpoint_config != shared.cmd_opts.config:
print(f"Loading config from: {checkpoint_config}")
if shared.sd_model:
sd_hijack.model_hijack.undo_hijack(shared.sd_model)
@ -289,7 +314,7 @@ def load_model(checkpoint_info=None):
gc.collect()
devices.torch_gc()
sd_config = OmegaConf.load(checkpoint_info.config)
sd_config = OmegaConf.load(checkpoint_config)
if should_hijack_inpainting(checkpoint_info):
# Hardcoded config for now...
@ -298,9 +323,6 @@ def load_model(checkpoint_info=None):
sd_config.model.params.unet_config.params.in_channels = 9
sd_config.model.params.finetune_keys = None
# Create a "fake" config with a different name so that we know to unload it when switching models.
checkpoint_info = checkpoint_info._replace(config=checkpoint_info.config.replace(".yaml", "-inpainting.yaml"))
if not hasattr(sd_config.model.params, "use_ema"):
sd_config.model.params.use_ema = False
@ -309,9 +331,21 @@ def load_model(checkpoint_info=None):
if shared.cmd_opts.no_half:
sd_config.model.params.unet_config.params.use_fp16 = False
sd_model = instantiate_from_config(sd_config.model)
timer = Timer()
try:
with sd_disable_initialization.DisableInitialization():
sd_model = instantiate_from_config(sd_config.model)
except Exception as e:
print('Failed to create model quickly; will retry using slow method.', file=sys.stderr)
sd_model = instantiate_from_config(sd_config.model)
elapsed_create = timer.elapsed()
load_model_weights(sd_model, checkpoint_info)
elapsed_load_weights = timer.elapsed()
if shared.cmd_opts.lowvram or shared.cmd_opts.medvram:
lowvram.setup_for_low_vram(sd_model, shared.cmd_opts.medvram)
else:
@ -322,11 +356,13 @@ def load_model(checkpoint_info=None):
sd_model.eval()
shared.sd_model = sd_model
sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings(force_reload=True) # Reload embeddings after model load as they may or may not fit the model
script_callbacks.model_loaded_callback(sd_model)
print("Model loaded.")
elapsed_the_rest = timer.elapsed()
sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings(force_reload = True) # Reload embeddings after model load as they may or may not fit the model
print(f"Model loaded in {elapsed_create + elapsed_load_weights + elapsed_the_rest:.1f}s ({elapsed_create:.1f}s create model, {elapsed_load_weights:.1f}s load weights).")
return sd_model
@ -334,14 +370,19 @@ def load_model(checkpoint_info=None):
def reload_model_weights(sd_model=None, info=None):
from modules import lowvram, devices, sd_hijack
checkpoint_info = info or select_checkpoint()
if not sd_model:
sd_model = shared.sd_model
if sd_model is None: # previous model load failed
current_checkpoint_info = None
else:
current_checkpoint_info = sd_model.sd_checkpoint_info
if sd_model.sd_model_checkpoint == checkpoint_info.filename:
return
if sd_model.sd_model_checkpoint == checkpoint_info.filename:
return
checkpoint_config = find_checkpoint_config(current_checkpoint_info)
if sd_model.sd_checkpoint_info.config != checkpoint_info.config or should_hijack_inpainting(checkpoint_info) != should_hijack_inpainting(sd_model.sd_checkpoint_info):
if current_checkpoint_info is None or checkpoint_config != find_checkpoint_config(checkpoint_info) or should_hijack_inpainting(checkpoint_info) != should_hijack_inpainting(sd_model.sd_checkpoint_info):
del sd_model
checkpoints_loaded.clear()
load_model(checkpoint_info)
@ -354,13 +395,23 @@ def reload_model_weights(sd_model=None, info=None):
sd_hijack.model_hijack.undo_hijack(sd_model)
load_model_weights(sd_model, checkpoint_info)
timer = Timer()
sd_hijack.model_hijack.hijack(sd_model)
script_callbacks.model_loaded_callback(sd_model)
try:
load_model_weights(sd_model, checkpoint_info)
except Exception as e:
print("Failed to load checkpoint, restoring previous")
load_model_weights(sd_model, current_checkpoint_info)
raise
finally:
sd_hijack.model_hijack.hijack(sd_model)
script_callbacks.model_loaded_callback(sd_model)
if not shared.cmd_opts.lowvram and not shared.cmd_opts.medvram:
sd_model.to(devices.device)
if not shared.cmd_opts.lowvram and not shared.cmd_opts.medvram:
sd_model.to(devices.device)
elapsed = timer.elapsed()
print(f"Weights loaded in {elapsed:.1f}s.")
print("Weights loaded.")
return sd_model

14
modules/sd_samplers.py
View File

@ -97,8 +97,9 @@ sampler_extra_params = {
def setup_img2img_steps(p, steps=None):
if opts.img2img_fix_steps or steps is not None:
steps = int((steps or p.steps) / min(p.denoising_strength, 0.999)) if p.denoising_strength > 0 else 0
t_enc = p.steps - 1
requested_steps = (steps or p.steps)
steps = int(requested_steps / min(p.denoising_strength, 0.999)) if p.denoising_strength > 0 else 0
t_enc = requested_steps - 1
else:
steps = p.steps
t_enc = int(min(p.denoising_strength, 0.999) * steps)
@ -462,6 +463,13 @@ class KDiffusionSampler:
return extra_params_kwargs
def get_sigmas(self, p, steps):
discard_next_to_last_sigma = self.config is not None and self.config.options.get('discard_next_to_last_sigma', False)
if opts.always_discard_next_to_last_sigma and not discard_next_to_last_sigma:
discard_next_to_last_sigma = True
p.extra_generation_params["Discard penultimate sigma"] = True
steps += 1 if discard_next_to_last_sigma else 0
if p.sampler_noise_scheduler_override:
sigmas = p.sampler_noise_scheduler_override(steps)
elif self.config is not None and self.config.options.get('scheduler', None) == 'karras':
@ -471,7 +479,7 @@ class KDiffusionSampler:
else:
sigmas = self.model_wrap.get_sigmas(steps)
if self.config is not None and self.config.options.get('discard_next_to_last_sigma', False):
if discard_next_to_last_sigma:
sigmas = torch.cat([sigmas[:-2], sigmas[-1:]])
return sigmas

20
modules/sd_vae.py
View File

@ -1,8 +1,9 @@
import torch
import safetensors.torch
import os
import collections
from collections import namedtuple
from modules import shared, devices, script_callbacks
from modules import shared, devices, script_callbacks, sd_models
from modules.paths import models_path
import glob
from copy import deepcopy
@ -72,8 +73,10 @@ def refresh_vae_list(vae_path=vae_path, model_path=model_path):
candidates = [
*glob.iglob(os.path.join(model_path, '**/*.vae.ckpt'), recursive=True),
*glob.iglob(os.path.join(model_path, '**/*.vae.pt'), recursive=True),
*glob.iglob(os.path.join(model_path, '**/*.vae.safetensors'), recursive=True),
*glob.iglob(os.path.join(vae_path, '**/*.ckpt'), recursive=True),
*glob.iglob(os.path.join(vae_path, '**/*.pt'), recursive=True)
*glob.iglob(os.path.join(vae_path, '**/*.pt'), recursive=True),
*glob.iglob(os.path.join(vae_path, '**/*.safetensors'), recursive=True),
]
if shared.cmd_opts.vae_path is not None and os.path.isfile(shared.cmd_opts.vae_path):
candidates.append(shared.cmd_opts.vae_path)
@ -137,6 +140,12 @@ def resolve_vae(checkpoint_file=None, vae_file="auto"):
if os.path.isfile(vae_file_try):
vae_file = vae_file_try
print(f"Using VAE found similar to selected model: {vae_file}")
# if still not found, try look for ".vae.safetensors" beside model
if vae_file == "auto":
vae_file_try = model_path + ".vae.safetensors"
if os.path.isfile(vae_file_try):
vae_file = vae_file_try
print(f"Using VAE found similar to selected model: {vae_file}")
# No more fallbacks for auto
if vae_file == "auto":
vae_file = None
@ -163,8 +172,9 @@ def load_vae(model, vae_file=None):
assert os.path.isfile(vae_file), f"VAE file doesn't exist: {vae_file}"
print(f"Loading VAE weights from: {vae_file}")
store_base_vae(model)
vae_ckpt = torch.load(vae_file, map_location=shared.weight_load_location)
vae_dict_1 = {k: v for k, v in vae_ckpt["state_dict"].items() if k[0:4] != "loss" and k not in vae_ignore_keys}
vae_ckpt = sd_models.read_state_dict(vae_file, map_location=shared.weight_load_location)
vae_dict_1 = {k: v for k, v in vae_ckpt.items() if k[0:4] != "loss" and k not in vae_ignore_keys}
_load_vae_dict(model, vae_dict_1)
if cache_enabled:
@ -195,10 +205,12 @@ def _load_vae_dict(model, vae_dict_1):
model.first_stage_model.load_state_dict(vae_dict_1)
model.first_stage_model.to(devices.dtype_vae)
def clear_loaded_vae():
global loaded_vae_file
loaded_vae_file = None
def reload_vae_weights(sd_model=None, vae_file="auto"):
from modules import lowvram, devices, sd_hijack

57
modules/shared.py
View File

@ -14,7 +14,7 @@ import modules.interrogate
import modules.memmon
import modules.styles
import modules.devices as devices
from modules import localization, sd_vae, extensions, script_loading
from modules import localization, sd_vae, extensions, script_loading, errors, ui_components
from modules.paths import models_path, script_path, sd_path
@ -33,6 +33,7 @@ parser.add_argument("--no-half-vae", action='store_true', help="do not switch th
parser.add_argument("--no-progressbar-hiding", action='store_true', help="do not hide progressbar in gradio UI (we hide it because it slows down ML if you have hardware acceleration in browser)")
parser.add_argument("--max-batch-count", type=int, default=16, help="maximum batch count value for the UI")
parser.add_argument("--embeddings-dir", type=str, default=os.path.join(script_path, 'embeddings'), help="embeddings directory for textual inversion (default: embeddings)")
parser.add_argument("--textual-inversion-templates-dir", type=str, default=os.path.join(script_path, 'textual_inversion_templates'), help="directory with textual inversion templates")
parser.add_argument("--hypernetwork-dir", type=str, default=os.path.join(models_path, 'hypernetworks'), help="hypernetwork directory")
parser.add_argument("--localizations-dir", type=str, default=os.path.join(script_path, 'localizations'), help="localizations directory")
parser.add_argument("--allow-code", action='store_true', help="allow custom script execution from webui")
@ -56,6 +57,10 @@ parser.add_argument("--xformers", action='store_true', help="enable xformers for
parser.add_argument("--force-enable-xformers", action='store_true', help="enable xformers for cross attention layers regardless of whether the checking code thinks you can run it; do not make bug reports if this fails to work")
parser.add_argument("--deepdanbooru", action='store_true', help="does not do anything")
parser.add_argument("--opt-split-attention", action='store_true', help="force-enables Doggettx's cross-attention layer optimization. By default, it's on for torch cuda.")
parser.add_argument("--opt-sub-quad-attention", action='store_true', help="enable memory efficient sub-quadratic cross-attention layer optimization")
parser.add_argument("--sub-quad-q-chunk-size", type=int, help="query chunk size for the sub-quadratic cross-attention layer optimization to use", default=1024)
parser.add_argument("--sub-quad-kv-chunk-size", type=int, help="kv chunk size for the sub-quadratic cross-attention layer optimization to use", default=None)
parser.add_argument("--sub-quad-chunk-threshold", type=int, help="the percentage of VRAM threshold for the sub-quadratic cross-attention layer optimization to use chunking", default=None)
parser.add_argument("--opt-split-attention-invokeai", action='store_true', help="force-enables InvokeAI's cross-attention layer optimization. By default, it's on when cuda is unavailable.")
parser.add_argument("--opt-split-attention-v1", action='store_true', help="enable older version of split attention optimization that does not consume all the VRAM it can find")
parser.add_argument("--disable-opt-split-attention", action='store_true', help="force-disables cross-attention layer optimization")
@ -82,6 +87,7 @@ parser.add_argument('--vae-path', type=str, help='Path to Variational Autoencode
parser.add_argument("--disable-safe-unpickle", action='store_true', help="disable checking pytorch models for malicious code", default=False)
parser.add_argument("--api", action='store_true', help="use api=True to launch the API together with the webui (use --nowebui instead for only the API)")
parser.add_argument("--api-auth", type=str, help='Set authentication for API like "username:password"; or comma-delimit multiple like "u1:p1,u2:p2,u3:p3"', default=None)
parser.add_argument("--api-log", action='store_true', help="use api-log=True to enable logging of all API requests")
parser.add_argument("--nowebui", action='store_true', help="use api=True to launch the API instead of the webui")
parser.add_argument("--ui-debug-mode", action='store_true', help="Don't load model to quickly launch UI")
parser.add_argument("--device-id", type=str, help="Select the default CUDA device to use (export CUDA_VISIBLE_DEVICES=0,1,etc might be needed before)", default=None)
@ -109,6 +115,17 @@ restricted_opts = {
"outdir_save",
}
ui_reorder_categories = [
"sampler",
"dimensions",
"cfg",
"seed",
"checkboxes",
"hires_fix",
"batch",
"scripts",
]
cmd_opts.disable_extension_access = (cmd_opts.share or cmd_opts.listen or cmd_opts.server_name) and not cmd_opts.enable_insecure_extension_access
devices.device, devices.device_interrogate, devices.device_gfpgan, devices.device_esrgan, devices.device_codeformer = \
@ -141,6 +158,7 @@ class State:
job = ""
job_no = 0
job_count = 0
processing_has_refined_job_count = False
job_timestamp = '0'
sampling_step = 0
sampling_steps = 0
@ -171,6 +189,7 @@ class State:
"interrupted": self.interrupted,
"job": self.job,
"job_count": self.job_count,
"job_timestamp": self.job_timestamp,
"job_no": self.job_no,
"sampling_step": self.sampling_step,
"sampling_steps": self.sampling_steps,
@ -181,6 +200,7 @@ class State:
def begin(self):
self.sampling_step = 0
self.job_count = -1
self.processing_has_refined_job_count = False
self.job_no = 0
self.job_timestamp = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
self.current_latent = None
@ -201,12 +221,13 @@ class State:
"""sets self.current_image from self.current_latent if enough sampling steps have been made after the last call to this"""
def set_current_image(self):
if not parallel_processing_allowed:
return
if self.sampling_step - self.current_image_sampling_step >= opts.show_progress_every_n_steps and opts.show_progress_every_n_steps > 0:
self.do_set_current_image()
def do_set_current_image(self):
if not parallel_processing_allowed:
return
if self.current_latent is None:
return
@ -218,6 +239,7 @@ class State:
self.current_image_sampling_step = self.sampling_step
state = State()
artist_db = modules.artists.ArtistsDatabase(os.path.join(script_path, 'artists.csv'))
@ -327,7 +349,6 @@ options_templates.update(options_section(('upscaling', "Upscaling"), {
"ESRGAN_tile_overlap": OptionInfo(8, "Tile overlap, in pixels for ESRGAN upscalers. Low values = visible seam.", gr.Slider, {"minimum": 0, "maximum": 48, "step": 1}),
"realesrgan_enabled_models": OptionInfo(["R-ESRGAN 4x+", "R-ESRGAN 4x+ Anime6B"], "Select which Real-ESRGAN models to show in the web UI. (Requires restart)", gr.CheckboxGroup, lambda: {"choices": realesrgan_models_names()}),
"upscaler_for_img2img": OptionInfo(None, "Upscaler for img2img", gr.Dropdown, lambda: {"choices": [x.name for x in sd_upscalers]}),
"use_scale_latent_for_hires_fix": OptionInfo(False, "Upscale latent space image when doing hires. fix"),
}))
options_templates.update(options_section(('face-restoration', "Face restoration"), {
@ -345,7 +366,8 @@ options_templates.update(options_section(('system', "System"), {
options_templates.update(options_section(('training', "Training"), {
"unload_models_when_training": OptionInfo(False, "Move VAE and CLIP to RAM when training if possible. Saves VRAM."),
"pin_memory": OptionInfo(False, "Turn on pin_memory for DataLoader. Makes training slightly faster but can increase memory usage."),
"save_optimizer_state": OptionInfo(False, "Saves Optimizer state as separate *.optim file. Training can be resumed with HN itself and matching optim file."),
"save_optimizer_state": OptionInfo(False, "Saves Optimizer state as separate *.optim file. Training of embedding or HN can be resumed with the matching optim file."),
"save_training_settings_to_txt": OptionInfo(True, "Save textual inversion and hypernet settings to a text file whenever training starts."),
"dataset_filename_word_regex": OptionInfo("", "Filename word regex"),
"dataset_filename_join_string": OptionInfo(" ", "Filename join string"),
"training_image_repeats_per_epoch": OptionInfo(1, "Number of repeats for a single input image per epoch; used only for displaying epoch number", gr.Number, {"precision": 0}),
@ -365,7 +387,7 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), {
"initial_noise_multiplier": OptionInfo(1.0, "Noise multiplier for img2img", gr.Slider, {"minimum": 0.5, "maximum": 1.5, "step": 0.01 }),
"img2img_color_correction": OptionInfo(False, "Apply color correction to img2img results to match original colors."),
"img2img_fix_steps": OptionInfo(False, "With img2img, do exactly the amount of steps the slider specifies (normally you'd do less with less denoising)."),
"img2img_background_color": OptionInfo("#ffffff", "With img2img, fill image's transparent parts with this color.", gr.ColorPicker, {}),
"img2img_background_color": OptionInfo("#ffffff", "With img2img, fill image's transparent parts with this color.", ui_components.FormColorPicker, {}),
"enable_quantization": OptionInfo(False, "Enable quantization in K samplers for sharper and cleaner results. This may change existing seeds. Requires restart to apply."),
"enable_emphasis": OptionInfo(True, "Emphasis: use (text) to make model pay more attention to text and [text] to make it pay less attention"),
"enable_batch_seeds": OptionInfo(True, "Make K-diffusion samplers produce same images in a batch as when making a single image"),
@ -377,6 +399,7 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), {
options_templates.update(options_section(('compatibility', "Compatibility"), {
"use_old_emphasis_implementation": OptionInfo(False, "Use old emphasis implementation. Can be useful to reproduce old seeds."),
"use_old_karras_scheduler_sigmas": OptionInfo(False, "Use old karras scheduler sigmas (0.1 to 10)."),
"use_old_hires_fix_width_height": OptionInfo(False, "For hires fix, use width/height sliders to set final resolution rather than first pass (disables Upscale by, Resize width/height to)."),
}))
options_templates.update(options_section(('interrogate', "Interrogate Options"), {
@ -411,7 +434,9 @@ options_templates.update(options_section(('ui', "User interface"), {
"js_modal_lightbox_initially_zoomed": OptionInfo(True, "Show images zoomed in by default in full page image viewer"),
"show_progress_in_title": OptionInfo(True, "Show generation progress in window title."),
"samplers_in_dropdown": OptionInfo(True, "Use dropdown for sampler selection instead of radio group"),
"dimensions_and_batch_together": OptionInfo(True, "Show Witdth/Height and Batch sliders in same row"),
'quicksettings': OptionInfo("sd_model_checkpoint", "Quicksettings list"),
'ui_reorder': OptionInfo(", ".join(ui_reorder_categories), "txt2img/img2img UI item order"),
'localization': OptionInfo("None", "Localization (requires restart)", gr.Dropdown, lambda: {"choices": ["None"] + list(localization.localizations.keys())}, refresh=lambda: localization.list_localizations(cmd_opts.localizations_dir)),
}))
@ -424,6 +449,7 @@ options_templates.update(options_section(('sampler-params', "Sampler parameters"
's_tmin': OptionInfo(0.0, "sigma tmin", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
's_noise': OptionInfo(1.0, "sigma noise", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
'eta_noise_seed_delta': OptionInfo(0, "Eta noise seed delta", gr.Number, {"precision": 0}),
'always_discard_next_to_last_sigma': OptionInfo(False, "Always discard next-to-last sigma"),
}))
options_templates.update(options_section((None, "Hidden options"), {
@ -482,7 +508,12 @@ class Options:
return False
if self.data_labels[key].onchange is not None:
self.data_labels[key].onchange()
try:
self.data_labels[key].onchange()
except Exception as e:
errors.display(e, f"changing setting {key} to {value}")
setattr(self, key, oldval)
return False
return True
@ -545,6 +576,16 @@ opts = Options()
if os.path.exists(config_filename):
opts.load(config_filename)
latent_upscale_default_mode = "Latent"
latent_upscale_modes = {
"Latent": {"mode": "bilinear", "antialias": False},
"Latent (antialiased)": {"mode": "bilinear", "antialias": True},
"Latent (bicubic)": {"mode": "bicubic", "antialias": False},
"Latent (bicubic antialiased)": {"mode": "bicubic", "antialias": True},
"Latent (nearest)": {"mode": "nearest", "antialias": False},
"Latent (nearest-exact)": {"mode": "nearest-exact", "antialias": False},
}
sd_upscalers = []
sd_model = None
@ -578,7 +619,7 @@ class TotalTQDM:
return
if self._tqdm is None:
self.reset()
self._tqdm.total=new_total
self._tqdm.total = new_total
def clear(self):
if self._tqdm is not None:

214
modules/sub_quadratic_attention.py Normal file
View File

@ -0,0 +1,214 @@
# original source:
# https://github.com/AminRezaei0x443/memory-efficient-attention/blob/1bc0d9e6ac5f82ea43a375135c4e1d3896ee1694/memory_efficient_attention/attention_torch.py
# license:
# MIT License (see Memory Efficient Attention under the Licenses section in the web UI interface for the full license)
# credit:
# Amin Rezaei (original author)
# Alex Birch (optimized algorithm for 3D tensors, at the expense of removing bias, masking and callbacks)
# brkirch (modified to use torch.narrow instead of dynamic_slice implementation)
# implementation of:
# Self-attention Does Not Need O(n2) Memory":
# https://arxiv.org/abs/2112.05682v2
from functools import partial
import torch
from torch import Tensor
from torch.utils.checkpoint import checkpoint
import math
from typing import Optional, NamedTuple, List
def narrow_trunc(
input: Tensor,
dim: int,
start: int,
length: int
) -> Tensor:
return torch.narrow(input, dim, start, length if input.shape[dim] >= start + length else input.shape[dim] - start)
class AttnChunk(NamedTuple):
exp_values: Tensor
exp_weights_sum: Tensor
max_score: Tensor
class SummarizeChunk:
@staticmethod
def __call__(
query: Tensor,
key: Tensor,
value: Tensor,
) -> AttnChunk: ...
class ComputeQueryChunkAttn:
@staticmethod
def __call__(
query: Tensor,
key: Tensor,
value: Tensor,
) -> Tensor: ...
def _summarize_chunk(
query: Tensor,
key: Tensor,
value: Tensor,
scale: float,
) -> AttnChunk:
attn_weights = torch.baddbmm(
torch.empty(1, 1, 1, device=query.device, dtype=query.dtype),
query,
key.transpose(1,2),
alpha=scale,
beta=0,
)
max_score, _ = torch.max(attn_weights, -1, keepdim=True)
max_score = max_score.detach()
exp_weights = torch.exp(attn_weights - max_score)
exp_values = torch.bmm(exp_weights, value)
max_score = max_score.squeeze(-1)
return AttnChunk(exp_values, exp_weights.sum(dim=-1), max_score)
def _query_chunk_attention(
query: Tensor,
key: Tensor,
value: Tensor,
summarize_chunk: SummarizeChunk,
kv_chunk_size: int,
) -> Tensor:
batch_x_heads, k_tokens, k_channels_per_head = key.shape
_, _, v_channels_per_head = value.shape
def chunk_scanner(chunk_idx: int) -> AttnChunk:
key_chunk = narrow_trunc(
key,
1,
chunk_idx,
kv_chunk_size
)
value_chunk = narrow_trunc(
value,
1,
chunk_idx,
kv_chunk_size
)
return summarize_chunk(query, key_chunk, value_chunk)
chunks: List[AttnChunk] = [
chunk_scanner(chunk) for chunk in torch.arange(0, k_tokens, kv_chunk_size)
]
acc_chunk = AttnChunk(*map(torch.stack, zip(*chunks)))
chunk_values, chunk_weights, chunk_max = acc_chunk
global_max, _ = torch.max(chunk_max, 0, keepdim=True)
max_diffs = torch.exp(chunk_max - global_max)
chunk_values *= torch.unsqueeze(max_diffs, -1)
chunk_weights *= max_diffs
all_values = chunk_values.sum(dim=0)
all_weights = torch.unsqueeze(chunk_weights, -1).sum(dim=0)
return all_values / all_weights
# TODO: refactor CrossAttention#get_attention_scores to share code with this
def _get_attention_scores_no_kv_chunking(
query: Tensor,
key: Tensor,
value: Tensor,
scale: float,
) -> Tensor:
attn_scores = torch.baddbmm(
torch.empty(1, 1, 1, device=query.device, dtype=query.dtype),
query,
key.transpose(1,2),
alpha=scale,
beta=0,
)
attn_probs = attn_scores.softmax(dim=-1)
del attn_scores
hidden_states_slice = torch.bmm(attn_probs, value)
return hidden_states_slice
class ScannedChunk(NamedTuple):
chunk_idx: int
attn_chunk: AttnChunk
def efficient_dot_product_attention(
query: Tensor,
key: Tensor,
value: Tensor,
query_chunk_size=1024,
kv_chunk_size: Optional[int] = None,
kv_chunk_size_min: Optional[int] = None,
use_checkpoint=True,
):
"""Computes efficient dot-product attention given query, key, and value.
This is efficient version of attention presented in
https://arxiv.org/abs/2112.05682v2 which comes with O(sqrt(n)) memory requirements.
Args:
query: queries for calculating attention with shape of
`[batch * num_heads, tokens, channels_per_head]`.
key: keys for calculating attention with shape of
`[batch * num_heads, tokens, channels_per_head]`.
value: values to be used in attention with shape of
`[batch * num_heads, tokens, channels_per_head]`.
query_chunk_size: int: query chunks size
kv_chunk_size: Optional[int]: key/value chunks size. if None: defaults to sqrt(key_tokens)
kv_chunk_size_min: Optional[int]: key/value minimum chunk size. only considered when kv_chunk_size is None. changes `sqrt(key_tokens)` into `max(sqrt(key_tokens), kv_chunk_size_min)`, to ensure our chunk sizes don't get too small (smaller chunks = more chunks = less concurrent work done).
use_checkpoint: bool: whether to use checkpointing (recommended True for training, False for inference)
Returns:
Output of shape `[batch * num_heads, query_tokens, channels_per_head]`.
"""
batch_x_heads, q_tokens, q_channels_per_head = query.shape
_, k_tokens, _ = key.shape
scale = q_channels_per_head ** -0.5
kv_chunk_size = min(kv_chunk_size or int(math.sqrt(k_tokens)), k_tokens)
if kv_chunk_size_min is not None:
kv_chunk_size = max(kv_chunk_size, kv_chunk_size_min)
def get_query_chunk(chunk_idx: int) -> Tensor:
return narrow_trunc(
query,
1,
chunk_idx,
min(query_chunk_size, q_tokens)
)
summarize_chunk: SummarizeChunk = partial(_summarize_chunk, scale=scale)
summarize_chunk: SummarizeChunk = partial(checkpoint, summarize_chunk) if use_checkpoint else summarize_chunk
compute_query_chunk_attn: ComputeQueryChunkAttn = partial(
_get_attention_scores_no_kv_chunking,
scale=scale
) if k_tokens <= kv_chunk_size else (
# fast-path for when there's just 1 key-value chunk per query chunk (this is just sliced attention btw)
partial(
_query_chunk_attention,
kv_chunk_size=kv_chunk_size,
summarize_chunk=summarize_chunk,
)
)
if q_tokens <= query_chunk_size:
# fast-path for when there's just 1 query chunk
return compute_query_chunk_attn(
query=query,
key=key,
value=value,
)
# TODO: maybe we should use torch.empty_like(query) to allocate storage in-advance,
# and pass slices to be mutated, instead of torch.cat()ing the returned slices
res = torch.cat([
compute_query_chunk_attn(
query=get_query_chunk(i * query_chunk_size),
key=key,
value=value,
) for i in range(math.ceil(q_tokens / query_chunk_size))
], dim=1)
return res

10
modules/textual_inversion/dataset.py
View File

@ -28,13 +28,11 @@ class DatasetEntry:
class PersonalizedBase(Dataset):
def __init__(self, data_root, width, height, repeats, flip_p=0.5, placeholder_token="*", model=None, cond_model=None, device=None, template_file=None, include_cond=False, batch_size=1, gradient_step=1, shuffle_tags=False, tag_drop_out=0, latent_sampling_method='once'):
def __init__(self, data_root, width, height, repeats, flip_p=0.5, placeholder_token="*", model=None, cond_model=None, device=None, template_file=None, include_cond=False, batch_size=1, gradient_step=1, shuffle_tags=False, tag_drop_out=0, latent_sampling_method='once', varsize=False):
re_word = re.compile(shared.opts.dataset_filename_word_regex) if len(shared.opts.dataset_filename_word_regex) > 0 else None
self.placeholder_token = placeholder_token
self.width = width
self.height = height
self.flip = transforms.RandomHorizontalFlip(p=flip_p)
self.dataset = []
@ -47,10 +45,10 @@ class PersonalizedBase(Dataset):
assert data_root, 'dataset directory not specified'
assert os.path.isdir(data_root), "Dataset directory doesn't exist"
assert os.listdir(data_root), "Dataset directory is empty"
assert batch_size == 1 or not varsize, 'variable img size must have batch size 1'
self.image_paths = [os.path.join(data_root, file_path) for file_path in os.listdir(data_root)]
self.shuffle_tags = shuffle_tags
self.tag_drop_out = tag_drop_out
@ -59,7 +57,9 @@ class PersonalizedBase(Dataset):
if shared.state.interrupted:
raise Exception("interrupted")
try:
image = Image.open(path).convert('RGB').resize((self.width, self.height), PIL.Image.BICUBIC)
image = Image.open(path).convert('RGB')
if not varsize:
image = image.resize((width, height), PIL.Image.BICUBIC)
except Exception:
continue

11
modules/textual_inversion/learn_schedule.py
View File

@ -58,14 +58,19 @@ class LearnRateScheduler:
self.finished = False
def apply(self, optimizer, step_number):
def step(self, step_number):
if step_number < self.end_step:
return
return False
try:
(self.learn_rate, self.end_step) = next(self.schedules)
except Exception:
except StopIteration:
self.finished = True
return False
return True
def apply(self, optimizer, step_number):
if not self.step(step_number):
return
if self.verbose:

24
modules/textual_inversion/logging.py Normal file
View File

@ -0,0 +1,24 @@
import datetime
import json
import os
saved_params_shared = {"model_name", "model_hash", "initial_step", "num_of_dataset_images", "learn_rate", "batch_size", "data_root", "log_directory", "training_width", "training_height", "steps", "create_image_every", "template_file"}
saved_params_ti = {"embedding_name", "num_vectors_per_token", "save_embedding_every", "save_image_with_stored_embedding"}
saved_params_hypernet = {"hypernetwork_name", "layer_structure", "activation_func", "weight_init", "add_layer_norm", "use_dropout", "save_hypernetwork_every"}
saved_params_all = saved_params_shared | saved_params_ti | saved_params_hypernet
saved_params_previews = {"preview_prompt", "preview_negative_prompt", "preview_steps", "preview_sampler_index", "preview_cfg_scale", "preview_seed", "preview_width", "preview_height"}
def save_settings_to_file(log_directory, all_params):
now = datetime.datetime.now()
params = {"datetime": now.strftime("%Y-%m-%d %H:%M:%S")}
keys = saved_params_all
if all_params.get('preview_from_txt2img'):
keys = keys | saved_params_previews
params.update({k: v for k, v in all_params.items() if k in keys})
filename = f'settings-{now.strftime("%Y-%m-%d-%H-%M-%S")}.json'
with open(os.path.join(log_directory, filename), "w") as file:
json.dump(params, file, indent=4)

1
modules/textual_inversion/preprocess.py
View File

@ -124,6 +124,7 @@ def preprocess_work(process_src, process_dst, process_width, process_height, pre
files = listfiles(src)
shared.state.job = "preprocess"
shared.state.textinfo = "Preprocessing..."
shared.state.job_count = len(files)

286
modules/textual_inversion/textual_inversion.py
View File

@ -1,6 +1,8 @@
import os
import sys
import traceback
import inspect
from collections import namedtuple
import torch
import tqdm
@ -14,9 +16,25 @@ from modules import shared, devices, sd_hijack, processing, sd_models, images, s
import modules.textual_inversion.dataset
from modules.textual_inversion.learn_schedule import LearnRateScheduler
from modules.textual_inversion.image_embedding import (embedding_to_b64, embedding_from_b64,
insert_image_data_embed, extract_image_data_embed,
caption_image_overlay)
from modules.textual_inversion.image_embedding import embedding_to_b64, embedding_from_b64, insert_image_data_embed, extract_image_data_embed, caption_image_overlay
from modules.textual_inversion.logging import save_settings_to_file
TextualInversionTemplate = namedtuple("TextualInversionTemplate", ["name", "path"])
textual_inversion_templates = {}
def list_textual_inversion_templates():
textual_inversion_templates.clear()
for root, dirs, fns in os.walk(shared.cmd_opts.textual_inversion_templates_dir):
for fn in fns:
path = os.path.join(root, fn)
textual_inversion_templates[fn] = TextualInversionTemplate(fn, path)
return textual_inversion_templates
class Embedding:
def __init__(self, vec, name, step=None):
@ -28,6 +46,7 @@ class Embedding:
self.cached_checksum = None
self.sd_checkpoint = None
self.sd_checkpoint_name = None
self.optimizer_state_dict = None
def save(self, filename):
embedding_data = {
@ -41,6 +60,13 @@ class Embedding:
torch.save(embedding_data, filename)
if shared.opts.save_optimizer_state and self.optimizer_state_dict is not None:
optimizer_saved_dict = {
'hash': self.checksum(),
'optimizer_state_dict': self.optimizer_state_dict,
}
torch.save(optimizer_saved_dict, filename + '.optim')
def checksum(self):
if self.cached_checksum is not None:
return self.cached_checksum
@ -55,20 +81,43 @@ class Embedding:
return self.cached_checksum
class DirWithTextualInversionEmbeddings:
def __init__(self, path):
self.path = path
self.mtime = None
def has_changed(self):
if not os.path.isdir(self.path):
return False
mt = os.path.getmtime(self.path)
if self.mtime is None or mt > self.mtime:
return True
def update(self):
if not os.path.isdir(self.path):
return
self.mtime = os.path.getmtime(self.path)
class EmbeddingDatabase:
def __init__(self, embeddings_dir):
def __init__(self):
self.ids_lookup = {}
self.word_embeddings = {}
self.skipped_embeddings = []
self.dir_mtime = None
self.embeddings_dir = embeddings_dir
self.skipped_embeddings = {}
self.expected_shape = -1
self.embedding_dirs = {}
def add_embedding_dir(self, path):
self.embedding_dirs[path] = DirWithTextualInversionEmbeddings(path)
def clear_embedding_dirs(self):
self.embedding_dirs.clear()
def register_embedding(self, embedding, model):
self.word_embeddings[embedding.name] = embedding
# TODO changing between clip and open clip changes tokenization, which will cause embeddings to stop working
ids = model.cond_stage_model.tokenize([embedding.name])[0]
first_id = ids[0]
@ -83,77 +132,98 @@ class EmbeddingDatabase:
vec = shared.sd_model.cond_stage_model.encode_embedding_init_text(",", 1)
return vec.shape[1]
def load_textual_inversion_embeddings(self, force_reload = False):
mt = os.path.getmtime(self.embeddings_dir)
if not force_reload and self.dir_mtime is not None and mt <= self.dir_mtime:
def load_from_file(self, path, filename):
name, ext = os.path.splitext(filename)
ext = ext.upper()
if ext in ['.PNG', '.WEBP', '.JXL', '.AVIF']:
_, second_ext = os.path.splitext(name)
if second_ext.upper() == '.PREVIEW':
return
embed_image = Image.open(path)
if hasattr(embed_image, 'text') and 'sd-ti-embedding' in embed_image.text:
data = embedding_from_b64(embed_image.text['sd-ti-embedding'])
name = data.get('name', name)
else:
data = extract_image_data_embed(embed_image)
name = data.get('name', name)
elif ext in ['.BIN', '.PT']:
data = torch.load(path, map_location="cpu")
else:
return
self.dir_mtime = mt
self.ids_lookup.clear()
self.word_embeddings.clear()
self.skipped_embeddings = []
self.expected_shape = self.get_expected_shape()
# textual inversion embeddings
if 'string_to_param' in data:
param_dict = data['string_to_param']
if hasattr(param_dict, '_parameters'):
param_dict = getattr(param_dict, '_parameters') # fix for torch 1.12.1 loading saved file from torch 1.11
assert len(param_dict) == 1, 'embedding file has multiple terms in it'
emb = next(iter(param_dict.items()))[1]
# diffuser concepts
elif type(data) == dict and type(next(iter(data.values()))) == torch.Tensor:
assert len(data.keys()) == 1, 'embedding file has multiple terms in it'
def process_file(path, filename):
name = os.path.splitext(filename)[0]
emb = next(iter(data.values()))
if len(emb.shape) == 1:
emb = emb.unsqueeze(0)
else:
raise Exception(f"Couldn't identify {filename} as neither textual inversion embedding nor diffuser concept.")
if os.path.splitext(filename.upper())[-1] in ['.PNG', '.WEBP', '.JXL', '.AVIF']:
embed_image = Image.open(path)
if hasattr(embed_image, 'text') and 'sd-ti-embedding' in embed_image.text:
data = embedding_from_b64(embed_image.text['sd-ti-embedding'])
name = data.get('name', name)
else:
data = extract_image_data_embed(embed_image)
name = data.get('name', name)
else:
data = torch.load(path, map_location="cpu")
vec = emb.detach().to(devices.device, dtype=torch.float32)
embedding = Embedding(vec, name)
embedding.step = data.get('step', None)
embedding.sd_checkpoint = data.get('sd_checkpoint', None)
embedding.sd_checkpoint_name = data.get('sd_checkpoint_name', None)
embedding.vectors = vec.shape[0]
embedding.shape = vec.shape[-1]
# textual inversion embeddings
if 'string_to_param' in data:
param_dict = data['string_to_param']
if hasattr(param_dict, '_parameters'):
param_dict = getattr(param_dict, '_parameters') # fix for torch 1.12.1 loading saved file from torch 1.11
assert len(param_dict) == 1, 'embedding file has multiple terms in it'
emb = next(iter(param_dict.items()))[1]
# diffuser concepts
elif type(data) == dict and type(next(iter(data.values()))) == torch.Tensor:
assert len(data.keys()) == 1, 'embedding file has multiple terms in it'
if self.expected_shape == -1 or self.expected_shape == embedding.shape:
self.register_embedding(embedding, shared.sd_model)
else:
self.skipped_embeddings[name] = embedding
emb = next(iter(data.values()))
if len(emb.shape) == 1:
emb = emb.unsqueeze(0)
else:
raise Exception(f"Couldn't identify {filename} as neither textual inversion embedding nor diffuser concept.")
def load_from_dir(self, embdir):
if not os.path.isdir(embdir.path):
return
vec = emb.detach().to(devices.device, dtype=torch.float32)
embedding = Embedding(vec, name)
embedding.step = data.get('step', None)
embedding.sd_checkpoint = data.get('sd_checkpoint', None)
embedding.sd_checkpoint_name = data.get('sd_checkpoint_name', None)
embedding.vectors = vec.shape[0]
embedding.shape = vec.shape[-1]
for root, dirs, fns in os.walk(embdir.path):
for fn in fns:
try:
fullfn = os.path.join(root, fn)
if self.expected_shape == -1 or self.expected_shape == embedding.shape:
self.register_embedding(embedding, shared.sd_model)
else:
self.skipped_embeddings.append(name)
if os.stat(fullfn).st_size == 0:
continue
for fn in os.listdir(self.embeddings_dir):
try:
fullfn = os.path.join(self.embeddings_dir, fn)
if os.stat(fullfn).st_size == 0:
self.load_from_file(fullfn, fn)
except Exception:
print(f"Error loading embedding {fn}:", file=sys.stderr)
print(traceback.format_exc(), file=sys.stderr)
continue
process_file(fullfn, fn)
except Exception:
print(f"Error loading embedding {fn}:", file=sys.stderr)
print(traceback.format_exc(), file=sys.stderr)
continue
def load_textual_inversion_embeddings(self, force_reload=False):
if not force_reload:
need_reload = False
for path, embdir in self.embedding_dirs.items():
if embdir.has_changed():
need_reload = True
break
if not need_reload:
return
self.ids_lookup.clear()
self.word_embeddings.clear()
self.skipped_embeddings.clear()
self.expected_shape = self.get_expected_shape()
for path, embdir in self.embedding_dirs.items():
self.load_from_dir(embdir)
embdir.update()
print(f"Textual inversion embeddings loaded({len(self.word_embeddings)}): {', '.join(self.word_embeddings.keys())}")
if len(self.skipped_embeddings) > 0:
print(f"Textual inversion embeddings skipped({len(self.skipped_embeddings)}): {', '.join(self.skipped_embeddings)}")
print(f"Textual inversion embeddings skipped({len(self.skipped_embeddings)}): {', '.join(self.skipped_embeddings.keys())}")
def find_embedding_at_position(self, tokens, offset):
token = tokens[offset]
@ -218,7 +288,8 @@ def write_loss(log_directory, filename, step, epoch_len, values):
**values,
})
def validate_train_inputs(model_name, learn_rate, batch_size, gradient_step, data_root, template_file, steps, save_model_every, create_image_every, log_directory, name="embedding"):
def validate_train_inputs(model_name, learn_rate, batch_size, gradient_step, data_root, template_file, template_filename, steps, save_model_every, create_image_every, log_directory, name="embedding"):
assert model_name, f"{name} not selected"
assert learn_rate, "Learning rate is empty or 0"
assert isinstance(batch_size, int), "Batch size must be integer"
@ -228,23 +299,28 @@ def validate_train_inputs(model_name, learn_rate, batch_size, gradient_step, dat
assert data_root, "Dataset directory is empty"
assert os.path.isdir(data_root), "Dataset directory doesn't exist"
assert os.listdir(data_root), "Dataset directory is empty"
assert template_file, "Prompt template file is empty"
assert os.path.isfile(template_file), "Prompt template file doesn't exist"
assert template_filename, "Prompt template file not selected"
assert template_file, f"Prompt template file {template_filename} not found"
assert os.path.isfile(template_file.path), f"Prompt template file {template_filename} doesn't exist"
assert steps, "Max steps is empty or 0"
assert isinstance(steps, int), "Max steps must be integer"
assert steps > 0 , "Max steps must be positive"
assert steps > 0, "Max steps must be positive"
assert isinstance(save_model_every, int), "Save {name} must be integer"
assert save_model_every >= 0 , "Save {name} must be positive or 0"
assert save_model_every >= 0, "Save {name} must be positive or 0"
assert isinstance(create_image_every, int), "Create image must be integer"
assert create_image_every >= 0 , "Create image must be positive or 0"
assert create_image_every >= 0, "Create image must be positive or 0"
if save_model_every or create_image_every:
assert log_directory, "Log directory is empty"
def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_root, log_directory, training_width, training_height, steps, shuffle_tags, tag_drop_out, latent_sampling_method, create_image_every, save_embedding_every, template_file, save_image_with_stored_embedding, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_root, log_directory, training_width, training_height, varsize, steps, clip_grad_mode, clip_grad_value, shuffle_tags, tag_drop_out, latent_sampling_method, create_image_every, save_embedding_every, template_filename, save_image_with_stored_embedding, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
save_embedding_every = save_embedding_every or 0
create_image_every = create_image_every or 0
validate_train_inputs(embedding_name, learn_rate, batch_size, gradient_step, data_root, template_file, steps, save_embedding_every, create_image_every, log_directory, name="embedding")
template_file = textual_inversion_templates.get(template_filename, None)
validate_train_inputs(embedding_name, learn_rate, batch_size, gradient_step, data_root, template_file, template_filename, steps, save_embedding_every, create_image_every, log_directory, name="embedding")
template_file = template_file.path
shared.state.job = "train-embedding"
shared.state.textinfo = "Initializing textual inversion training..."
shared.state.job_count = steps
@ -280,15 +356,23 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
if initial_step >= steps:
shared.state.textinfo = "Model has already been trained beyond specified max steps"
return embedding, filename
scheduler = LearnRateScheduler(learn_rate, steps, initial_step)
# dataset loading may take a while, so input validations and early returns should be done before this
clip_grad = torch.nn.utils.clip_grad_value_ if clip_grad_mode == "value" else \
torch.nn.utils.clip_grad_norm_ if clip_grad_mode == "norm" else \
None
if clip_grad:
clip_grad_sched = LearnRateScheduler(clip_grad_value, steps, initial_step, verbose=False)
# dataset loading may take a while, so input validations and early returns should be done before this
shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
old_parallel_processing_allowed = shared.parallel_processing_allowed
pin_memory = shared.opts.pin_memory
ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=embedding_name, model=shared.sd_model, cond_model=shared.sd_model.cond_stage_model, device=devices.device, template_file=template_file, batch_size=batch_size, gradient_step=gradient_step, shuffle_tags=shuffle_tags, tag_drop_out=tag_drop_out, latent_sampling_method=latent_sampling_method)
ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=embedding_name, model=shared.sd_model, cond_model=shared.sd_model.cond_stage_model, device=devices.device, template_file=template_file, batch_size=batch_size, gradient_step=gradient_step, shuffle_tags=shuffle_tags, tag_drop_out=tag_drop_out, latent_sampling_method=latent_sampling_method, varsize=varsize)
if shared.opts.save_training_settings_to_txt:
save_settings_to_file(log_directory, {**dict(model_name=checkpoint.model_name, model_hash=checkpoint.hash, num_of_dataset_images=len(ds), num_vectors_per_token=len(embedding.vec)), **locals()})
latent_sampling_method = ds.latent_sampling_method
@ -300,6 +384,19 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
embedding.vec.requires_grad = True
optimizer = torch.optim.AdamW([embedding.vec], lr=scheduler.learn_rate, weight_decay=0.0)
if shared.opts.save_optimizer_state:
optimizer_state_dict = None
if os.path.exists(filename + '.optim'):
optimizer_saved_dict = torch.load(filename + '.optim', map_location='cpu')
if embedding.checksum() == optimizer_saved_dict.get('hash', None):
optimizer_state_dict = optimizer_saved_dict.get('optimizer_state_dict', None)
if optimizer_state_dict is not None:
optimizer.load_state_dict(optimizer_state_dict)
print("Loaded existing optimizer from checkpoint")
else:
print("No saved optimizer exists in checkpoint")
scaler = torch.cuda.amp.GradScaler()
batch_size = ds.batch_size
@ -310,12 +407,14 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
loss_step = 0
_loss_step = 0 #internal
last_saved_file = "<none>"
last_saved_image = "<none>"
forced_filename = "<none>"
embedding_yet_to_be_embedded = False
is_training_inpainting_model = shared.sd_model.model.conditioning_key in {'hybrid', 'concat'}
img_c = None
pbar = tqdm.tqdm(total=steps - initial_step)
try:
for i in range((steps-initial_step) * gradient_step):
@ -333,14 +432,22 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
if shared.state.interrupted:
break
if clip_grad:
clip_grad_sched.step(embedding.step)
with devices.autocast():
# c = stack_conds(batch.cond).to(devices.device)
# mask = torch.tensor(batch.emb_index).to(devices.device, non_blocking=pin_memory)
# print(mask)
# c[:, 1:1+embedding.vec.shape[0]] = embedding.vec.to(devices.device, non_blocking=pin_memory)
x = batch.latent_sample.to(devices.device, non_blocking=pin_memory)
c = shared.sd_model.cond_stage_model(batch.cond_text)
loss = shared.sd_model(x, c)[0] / gradient_step
if is_training_inpainting_model:
if img_c is None:
img_c = processing.txt2img_image_conditioning(shared.sd_model, c, training_width, training_height)
cond = {"c_concat": [img_c], "c_crossattn": [c]}
else:
cond = c
loss = shared.sd_model(x, cond)[0] / gradient_step
del x
_loss_step += loss.item()
@ -349,6 +456,10 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
# go back until we reach gradient accumulation steps
if (j + 1) % gradient_step != 0:
continue
if clip_grad:
clip_grad(embedding.vec, clip_grad_sched.learn_rate)
scaler.step(optimizer)
scaler.update()
embedding.step += 1
@ -367,9 +478,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
# Before saving, change name to match current checkpoint.
embedding_name_every = f'{embedding_name}-{steps_done}'
last_saved_file = os.path.join(embedding_dir, f'{embedding_name_every}.pt')
#if shared.opts.save_optimizer_state:
#embedding.optimizer_state_dict = optimizer.state_dict()
save_embedding(embedding, checkpoint, embedding_name_every, last_saved_file, remove_cached_checksum=True)
save_embedding(embedding, optimizer, checkpoint, embedding_name_every, last_saved_file, remove_cached_checksum=True)
embedding_yet_to_be_embedded = True
write_loss(log_directory, "textual_inversion_loss.csv", embedding.step, steps_per_epoch, {
@ -459,7 +568,7 @@ Last saved image: {html.escape(last_saved_image)}<br/>
</p>
"""
filename = os.path.join(shared.cmd_opts.embeddings_dir, f'{embedding_name}.pt')
save_embedding(embedding, checkpoint, embedding_name, filename, remove_cached_checksum=True)
save_embedding(embedding, optimizer, checkpoint, embedding_name, filename, remove_cached_checksum=True)
except Exception:
print(traceback.format_exc(), file=sys.stderr)
pass
@ -471,7 +580,7 @@ Last saved image: {html.escape(last_saved_image)}<br/>
return embedding, filename
def save_embedding(embedding, checkpoint, embedding_name, filename, remove_cached_checksum=True):
def save_embedding(embedding, optimizer, checkpoint, embedding_name, filename, remove_cached_checksum=True):
old_embedding_name = embedding.name
old_sd_checkpoint = embedding.sd_checkpoint if hasattr(embedding, "sd_checkpoint") else None
old_sd_checkpoint_name = embedding.sd_checkpoint_name if hasattr(embedding, "sd_checkpoint_name") else None
@ -482,6 +591,7 @@ def save_embedding(embedding, checkpoint, embedding_name, filename, remove_cache
if remove_cached_checksum:
embedding.cached_checksum = None
embedding.name = embedding_name
embedding.optimizer_state_dict = optimizer.state_dict()
embedding.save(filename)
except:
embedding.sd_checkpoint = old_sd_checkpoint

9
modules/txt2img.py
View File

@ -8,7 +8,7 @@ import modules.processing as processing
from modules.ui import plaintext_to_html
def txt2img(prompt: str, negative_prompt: str, prompt_style: str, prompt_style2: str, steps: int, sampler_index: int, restore_faces: bool, tiling: bool, n_iter: int, batch_size: int, cfg_scale: float, seed: int, subseed: int, subseed_strength: float, seed_resize_from_h: int, seed_resize_from_w: int, seed_enable_extras: bool, height: int, width: int, enable_hr: bool, denoising_strength: float, firstphase_width: int, firstphase_height: int, *args):
def txt2img(prompt: str, negative_prompt: str, prompt_style: str, prompt_style2: str, steps: int, sampler_index: int, restore_faces: bool, tiling: bool, n_iter: int, batch_size: int, cfg_scale: float, seed: int, subseed: int, subseed_strength: float, seed_resize_from_h: int, seed_resize_from_w: int, seed_enable_extras: bool, height: int, width: int, enable_hr: bool, denoising_strength: float, hr_scale: float, hr_upscaler: str, hr_second_pass_steps: int, hr_resize_x: int, hr_resize_y: int, *args):
p = StableDiffusionProcessingTxt2Img(
sd_model=shared.sd_model,
outpath_samples=opts.outdir_samples or opts.outdir_txt2img_samples,
@ -33,8 +33,11 @@ def txt2img(prompt: str, negative_prompt: str, prompt_style: str, prompt_style2:
tiling=tiling,
enable_hr=enable_hr,
denoising_strength=denoising_strength if enable_hr else None,
firstphase_width=firstphase_width if enable_hr else None,
firstphase_height=firstphase_height if enable_hr else None,
hr_scale=hr_scale,
hr_upscaler=hr_upscaler,
hr_second_pass_steps=hr_second_pass_steps,
hr_resize_x=hr_resize_x,
hr_resize_y=hr_resize_y,
)
p.scripts = modules.scripts.scripts_txt2img

666
modules/ui.py
View File

File diff suppressed because it is too large Load Diff

21
modules/ui_components.py
View File

@ -16,3 +16,24 @@ class FormRow(gr.Row, gr.components.FormComponent):
def get_block_name(self):
return "row"
class FormGroup(gr.Group, gr.components.FormComponent):
"""Same as gr.Row but fits inside gradio forms"""
def get_block_name(self):
return "group"
class FormHTML(gr.HTML, gr.components.FormComponent):
"""Same as gr.HTML but fits inside gradio forms"""
def get_block_name(self):
return "html"
class FormColorPicker(gr.ColorPicker, gr.components.FormComponent):
"""Same as gr.ColorPicker but fits inside gradio forms"""
def get_block_name(self):
return "colorpicker"

44
modules/ui_extensions.py
View File

@ -162,15 +162,15 @@ def install_extension_from_url(dirname, url):
shutil.rmtree(tmpdir, True)
def install_extension_from_index(url, hide_tags):
def install_extension_from_index(url, hide_tags, sort_column):
ext_table, message = install_extension_from_url(None, url)
code, _ = refresh_available_extensions_from_data(hide_tags)
code, _ = refresh_available_extensions_from_data(hide_tags, sort_column)
return code, ext_table, message
def refresh_available_extensions(url, hide_tags):
def refresh_available_extensions(url, hide_tags, sort_column):
global available_extensions
import urllib.request
@ -179,18 +179,28 @@ def refresh_available_extensions(url, hide_tags):
available_extensions = json.loads(text)
code, tags = refresh_available_extensions_from_data(hide_tags)
code, tags = refresh_available_extensions_from_data(hide_tags, sort_column)
return url, code, gr.CheckboxGroup.update(choices=tags), ''
def refresh_available_extensions_for_tags(hide_tags):
code, _ = refresh_available_extensions_from_data(hide_tags)
def refresh_available_extensions_for_tags(hide_tags, sort_column):
code, _ = refresh_available_extensions_from_data(hide_tags, sort_column)
return code, ''
def refresh_available_extensions_from_data(hide_tags):
sort_ordering = [
# (reverse, order_by_function)
(True, lambda x: x.get('added', 'z')),
(False, lambda x: x.get('added', 'z')),
(False, lambda x: x.get('name', 'z')),
(True, lambda x: x.get('name', 'z')),
(False, lambda x: 'z'),
]
def refresh_available_extensions_from_data(hide_tags, sort_column):
extlist = available_extensions["extensions"]
installed_extension_urls = {normalize_git_url(extension.remote): extension.name for extension in extensions.extensions}
@ -210,8 +220,11 @@ def refresh_available_extensions_from_data(hide_tags):
<tbody>
"""
for ext in extlist:
sort_reverse, sort_function = sort_ordering[sort_column if 0 <= sort_column < len(sort_ordering) else 0]
for ext in sorted(extlist, key=sort_function, reverse=sort_reverse):
name = ext.get("name", "noname")
added = ext.get('added', 'unknown')
url = ext.get("url", None)
description = ext.get("description", "")
extension_tags = ext.get("tags", [])
@ -233,7 +246,7 @@ def refresh_available_extensions_from_data(hide_tags):
code += f"""
<tr>
<td><a href="{html.escape(url)}" target="_blank">{html.escape(name)}</a><br />{tags_text}</td>
<td>{html.escape(description)}</td>
<td>{html.escape(description)}<p class="info"><span class="date_added">Added: {html.escape(added)}</span></p></td>
<td>{install_code}</td>
</tr>
@ -291,25 +304,32 @@ def create_ui():
with gr.Row():
hide_tags = gr.CheckboxGroup(value=["ads", "localization", "installed"], label="Hide extensions with tags", choices=["script", "ads", "localization", "installed"])
sort_column = gr.Radio(value="newest first", label="Order", choices=["newest first", "oldest first", "a-z", "z-a", "internal order", ], type="index")
install_result = gr.HTML()
available_extensions_table = gr.HTML()
refresh_available_extensions_button.click(
fn=modules.ui.wrap_gradio_call(refresh_available_extensions, extra_outputs=[gr.update(), gr.update(), gr.update()]),
inputs=[available_extensions_index, hide_tags],
inputs=[available_extensions_index, hide_tags, sort_column],
outputs=[available_extensions_index, available_extensions_table, hide_tags, install_result],
)
install_extension_button.click(
fn=modules.ui.wrap_gradio_call(install_extension_from_index, extra_outputs=[gr.update(), gr.update()]),
inputs=[extension_to_install, hide_tags],
inputs=[extension_to_install, hide_tags, sort_column],
outputs=[available_extensions_table, extensions_table, install_result],
)
hide_tags.change(
fn=modules.ui.wrap_gradio_call(refresh_available_extensions_for_tags, extra_outputs=[gr.update()]),
inputs=[hide_tags],
inputs=[hide_tags, sort_column],
outputs=[available_extensions_table, install_result]
)
sort_column.change(
fn=modules.ui.wrap_gradio_call(refresh_available_extensions_for_tags, extra_outputs=[gr.update()]),
inputs=[hide_tags, sort_column],
outputs=[available_extensions_table, install_result]
)

101
modules/ui_progress.py Normal file
View File

@ -0,0 +1,101 @@
import time
import gradio as gr
from modules.shared import opts
import modules.shared as shared
def calc_time_left(progress, threshold, label, force_display, show_eta):
if progress == 0:
return ""
else:
time_since_start = time.time() - shared.state.time_start
eta = (time_since_start/progress)
eta_relative = eta-time_since_start
if (eta_relative > threshold and show_eta) or force_display:
if eta_relative > 3600:
return label + time.strftime('%H:%M:%S', time.gmtime(eta_relative))
elif eta_relative > 60:
return label + time.strftime('%M:%S', time.gmtime(eta_relative))
else:
return label + time.strftime('%Ss', time.gmtime(eta_relative))
else:
return ""
def check_progress_call(id_part):
if shared.state.job_count == 0:
return "", gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)
progress = 0
if shared.state.job_count > 0:
progress += shared.state.job_no / shared.state.job_count
if shared.state.sampling_steps > 0:
progress += 1 / shared.state.job_count * shared.state.sampling_step / shared.state.sampling_steps
# Show progress percentage and time left at the same moment, and base it also on steps done
show_eta = progress >= 0.01 or shared.state.sampling_step >= 10
time_left = calc_time_left(progress, 1, " ETA: ", shared.state.time_left_force_display, show_eta)
if time_left != "":
shared.state.time_left_force_display = True
progress = min(progress, 1)
progressbar = ""
if opts.show_progressbar:
progressbar = f"""<div class='progressDiv'><div class='progress' style="overflow:visible;width:{progress * 100}%;white-space:nowrap;">{"&nbsp;" * 2 + str(int(progress*100))+"%" + time_left if show_eta else ""}</div></div>"""
image = gr.update(visible=False)
preview_visibility = gr.update(visible=False)
if opts.show_progress_every_n_steps != 0:
shared.state.set_current_image()
image = shared.state.current_image
if image is None:
image = gr.update(value=None)
else:
preview_visibility = gr.update(visible=True)
if shared.state.textinfo is not None:
textinfo_result = gr.HTML.update(value=shared.state.textinfo, visible=True)
else:
textinfo_result = gr.update(visible=False)
return f"<span id='{id_part}_progress_span' style='display: none'>{time.time()}</span><p>{progressbar}</p>", preview_visibility, image, textinfo_result
def check_progress_call_initial(id_part):
shared.state.job_count = -1
shared.state.current_latent = None
shared.state.current_image = None
shared.state.textinfo = None
shared.state.time_start = time.time()
shared.state.time_left_force_display = False
return check_progress_call(id_part)
def setup_progressbar(progressbar, preview, id_part, textinfo=None):
if textinfo is None:
textinfo = gr.HTML(visible=False)
check_progress = gr.Button('Check progress', elem_id=f"{id_part}_check_progress", visible=False)
check_progress.click(
fn=lambda: check_progress_call(id_part),
show_progress=False,
inputs=[],
outputs=[progressbar, preview, preview, textinfo],
)
check_progress_initial = gr.Button('Check progress (first)', elem_id=f"{id_part}_check_progress_initial", visible=False)
check_progress_initial.click(
fn=lambda: check_progress_call_initial(id_part),
show_progress=False,
inputs=[],
outputs=[progressbar, preview, preview, textinfo],
)

23
modules/ui_tempdir.py
View File

@ -1,6 +1,7 @@
import os
import tempfile
from collections import namedtuple
from pathlib import Path
import gradio as gr
@ -12,10 +13,28 @@ from modules import shared
Savedfile = namedtuple("Savedfile", ["name"])
def register_tmp_file(gradio, filename):
if hasattr(gradio, 'temp_file_sets'): # gradio 3.15
gradio.temp_file_sets[0] = gradio.temp_file_sets[0] | {os.path.abspath(filename)}
if hasattr(gradio, 'temp_dirs'): # gradio 3.9
gradio.temp_dirs = gradio.temp_dirs | {os.path.abspath(os.path.dirname(filename))}
def check_tmp_file(gradio, filename):
if hasattr(gradio, 'temp_file_sets'):
return any([filename in fileset for fileset in gradio.temp_file_sets])
if hasattr(gradio, 'temp_dirs'):
return any(Path(temp_dir).resolve() in Path(filename).resolve().parents for temp_dir in gradio.temp_dirs)
return False
def save_pil_to_file(pil_image, dir=None):
already_saved_as = getattr(pil_image, 'already_saved_as', None)
if already_saved_as and os.path.isfile(already_saved_as):
shared.demo.temp_file_sets[0] = shared.demo.temp_file_sets[0] | {os.path.abspath(already_saved_as)}
register_tmp_file(shared.demo, already_saved_as)
file_obj = Savedfile(already_saved_as)
return file_obj
@ -45,7 +64,7 @@ def on_tmpdir_changed():
os.makedirs(shared.opts.temp_dir, exist_ok=True)
shared.demo.temp_file_sets[0] = shared.demo.temp_file_sets[0] | {os.path.abspath(shared.opts.temp_dir)}
register_tmp_file(shared.demo, os.path.join(shared.opts.temp_dir, "x"))
def cleanup_tmpdr():

6
modules/upscaler.py
View File

@ -53,10 +53,10 @@ class Upscaler:
def do_upscale(self, img: PIL.Image, selected_model: str):
return img
def upscale(self, img: PIL.Image, scale: int, selected_model: str = None):
def upscale(self, img: PIL.Image, scale, selected_model: str = None):
self.scale = scale
dest_w = img.width * scale
dest_h = img.height * scale
dest_w = int(img.width * scale)
dest_h = int(img.height * scale)
for i in range(3):
shape = (img.width, img.height)

4
requirements.txt
View File

@ -9,7 +9,7 @@ gradio==3.15.0
invisible-watermark
numpy
omegaconf
opencv-python
opencv-contrib-python
requests
piexif
Pillow
@ -30,4 +30,4 @@ inflection
GitPython
torchsde
safetensors
psutil; sys_platform == 'darwin'
psutil

4
requirements_versions.txt
View File

@ -5,7 +5,7 @@ basicsr==1.4.2
gfpgan==1.3.8
gradio==3.15.0
numpy==1.23.3
Pillow==9.2.0
Pillow==9.4.0
realesrgan==0.3.0
torch
omegaconf==2.2.3
@ -26,5 +26,5 @@ lark==1.1.2
inflection==0.5.1
GitPython==3.1.27
torchsde==0.2.5
safetensors==0.2.5
safetensors==0.2.7
httpcore<=0.15

BIN
screenshot.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 513 KiB

After

Width:  |  Height:  |  Size: 411 KiB

2
script.js
View File

@ -4,7 +4,7 @@ function gradioApp() {
}
function get_uiCurrentTab() {
return gradioApp().querySelector('.tabs button:not(.border-transparent)')
return gradioApp().querySelector('#tabs button:not(.border-transparent)')
}
function get_uiCurrentTabContent() {

3
scripts/custom_code.py
View File

@ -9,12 +9,11 @@ class Script(scripts.Script):
def title(self):
return "Custom code"
def show(self, is_img2img):
return cmd_opts.allow_code
def ui(self, is_img2img):
code = gr.Textbox(label="Python code", lines=1)
code = gr.Textbox(label="Python code", lines=1, elem_id=self.elem_id("code"))
return [code]

22
scripts/img2imgalt.py
View File

@ -125,25 +125,25 @@ class Script(scripts.Script):
def show(self, is_img2img):
return is_img2img
def ui(self, is_img2img):
def ui(self, is_img2img):
info = gr.Markdown('''
* `CFG Scale` should be 2 or lower.
''')
override_sampler = gr.Checkbox(label="Override `Sampling method` to Euler?(this method is built for it)", value=True)
override_sampler = gr.Checkbox(label="Override `Sampling method` to Euler?(this method is built for it)", value=True, elem_id=self.elem_id("override_sampler"))
override_prompt = gr.Checkbox(label="Override `prompt` to the same value as `original prompt`?(and `negative prompt`)", value=True)
original_prompt = gr.Textbox(label="Original prompt", lines=1)
original_negative_prompt = gr.Textbox(label="Original negative prompt", lines=1)
override_prompt = gr.Checkbox(label="Override `prompt` to the same value as `original prompt`?(and `negative prompt`)", value=True, elem_id=self.elem_id("override_prompt"))
original_prompt = gr.Textbox(label="Original prompt", lines=1, elem_id=self.elem_id("original_prompt"))
original_negative_prompt = gr.Textbox(label="Original negative prompt", lines=1, elem_id=self.elem_id("original_negative_prompt"))
override_steps = gr.Checkbox(label="Override `Sampling Steps` to the same value as `Decode steps`?", value=True)
st = gr.Slider(label="Decode steps", minimum=1, maximum=150, step=1, value=50)
override_steps = gr.Checkbox(label="Override `Sampling Steps` to the same value as `Decode steps`?", value=True, elem_id=self.elem_id("override_steps"))
st = gr.Slider(label="Decode steps", minimum=1, maximum=150, step=1, value=50, elem_id=self.elem_id("st"))
override_strength = gr.Checkbox(label="Override `Denoising strength` to 1?", value=True)
override_strength = gr.Checkbox(label="Override `Denoising strength` to 1?", value=True, elem_id=self.elem_id("override_strength"))
cfg = gr.Slider(label="Decode CFG scale", minimum=0.0, maximum=15.0, step=0.1, value=1.0)
randomness = gr.Slider(label="Randomness", minimum=0.0, maximum=1.0, step=0.01, value=0.0)
sigma_adjustment = gr.Checkbox(label="Sigma adjustment for finding noise for image", value=False)
cfg = gr.Slider(label="Decode CFG scale", minimum=0.0, maximum=15.0, step=0.1, value=1.0, elem_id=self.elem_id("cfg"))
randomness = gr.Slider(label="Randomness", minimum=0.0, maximum=1.0, step=0.01, value=0.0, elem_id=self.elem_id("randomness"))
sigma_adjustment = gr.Checkbox(label="Sigma adjustment for finding noise for image", value=False, elem_id=self.elem_id("sigma_adjustment"))
return [
info,

7
scripts/loopback.py
View File

@ -9,6 +9,7 @@ from modules.processing import Processed
from modules.sd_samplers import samplers
from modules.shared import opts, cmd_opts, state
class Script(scripts.Script):
def title(self):
return "Loopback"
@ -16,9 +17,9 @@ class Script(scripts.Script):
def show(self, is_img2img):
return is_img2img
def ui(self, is_img2img):
loops = gr.Slider(minimum=1, maximum=32, step=1, label='Loops', value=4)
denoising_strength_change_factor = gr.Slider(minimum=0.9, maximum=1.1, step=0.01, label='Denoising strength change factor', value=1)
def ui(self, is_img2img):
loops = gr.Slider(minimum=1, maximum=32, step=1, label='Loops', value=4, elem_id=self.elem_id("loops"))
denoising_strength_change_factor = gr.Slider(minimum=0.9, maximum=1.1, step=0.01, label='Denoising strength change factor', value=1, elem_id=self.elem_id("denoising_strength_change_factor"))
return [loops, denoising_strength_change_factor]

10
scripts/outpainting_mk_2.py
View File

@ -131,11 +131,11 @@ class Script(scripts.Script):
info = gr.HTML("<p style=\"margin-bottom:0.75em\">Recommended settings: Sampling Steps: 80-100, Sampler: Euler a, Denoising strength: 0.8</p>")
pixels = gr.Slider(label="Pixels to expand", minimum=8, maximum=256, step=8, value=128)
mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=8)
direction = gr.CheckboxGroup(label="Outpainting direction", choices=['left', 'right', 'up', 'down'], value=['left', 'right', 'up', 'down'])
noise_q = gr.Slider(label="Fall-off exponent (lower=higher detail)", minimum=0.0, maximum=4.0, step=0.01, value=1.0)
color_variation = gr.Slider(label="Color variation", minimum=0.0, maximum=1.0, step=0.01, value=0.05)
pixels = gr.Slider(label="Pixels to expand", minimum=8, maximum=256, step=8, value=128, elem_id=self.elem_id("pixels"))
mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=8, elem_id=self.elem_id("mask_blur"))
direction = gr.CheckboxGroup(label="Outpainting direction", choices=['left', 'right', 'up', 'down'], value=['left', 'right', 'up', 'down'], elem_id=self.elem_id("direction"))
noise_q = gr.Slider(label="Fall-off exponent (lower=higher detail)", minimum=0.0, maximum=4.0, step=0.01, value=1.0, elem_id=self.elem_id("noise_q"))
color_variation = gr.Slider(label="Color variation", minimum=0.0, maximum=1.0, step=0.01, value=0.05, elem_id=self.elem_id("color_variation"))
return [info, pixels, mask_blur, direction, noise_q, color_variation]

11
scripts/poor_mans_outpainting.py
View File

@ -9,7 +9,6 @@ from modules.processing import Processed, process_images
from modules.shared import opts, cmd_opts, state
class Script(scripts.Script):
def title(self):
return "Poor man's outpainting"
@ -20,11 +19,11 @@ class Script(scripts.Script):
def ui(self, is_img2img):
if not is_img2img:
return None
pixels = gr.Slider(label="Pixels to expand", minimum=8, maximum=256, step=8, value=128)
mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=4)
inpainting_fill = gr.Radio(label='Masked content', choices=['fill', 'original', 'latent noise', 'latent nothing'], value='fill', type="index")
direction = gr.CheckboxGroup(label="Outpainting direction", choices=['left', 'right', 'up', 'down'], value=['left', 'right', 'up', 'down'])
pixels = gr.Slider(label="Pixels to expand", minimum=8, maximum=256, step=8, value=128, elem_id=self.elem_id("pixels"))
mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=4, elem_id=self.elem_id("mask_blur"))
inpainting_fill = gr.Radio(label='Masked content', choices=['fill', 'original', 'latent noise', 'latent nothing'], value='fill', type="index", elem_id=self.elem_id("inpainting_fill"))
direction = gr.CheckboxGroup(label="Outpainting direction", choices=['left', 'right', 'up', 'down'], value=['left', 'right', 'up', 'down'], elem_id=self.elem_id("direction"))
return [pixels, mask_blur, inpainting_fill, direction]

6
scripts/prompt_matrix.py
View File

@ -44,9 +44,9 @@ class Script(scripts.Script):
def title(self):
return "Prompt matrix"
def ui(self, is_img2img):
put_at_start = gr.Checkbox(label='Put variable parts at start of prompt', value=False)
different_seeds = gr.Checkbox(label='Use different seed for each picture', value=False)
def ui(self, is_img2img):
put_at_start = gr.Checkbox(label='Put variable parts at start of prompt', value=False, elem_id=self.elem_id("put_at_start"))
different_seeds = gr.Checkbox(label='Use different seed for each picture', value=False, elem_id=self.elem_id("different_seeds"))
return [put_at_start, different_seeds]

10
scripts/prompts_from_file.py
View File

@ -111,12 +111,12 @@ class Script(scripts.Script):
def title(self):
return "Prompts from file or textbox"
def ui(self, is_img2img):
checkbox_iterate = gr.Checkbox(label="Iterate seed every line", value=False)
checkbox_iterate_batch = gr.Checkbox(label="Use same random seed for all lines", value=False)
def ui(self, is_img2img):
checkbox_iterate = gr.Checkbox(label="Iterate seed every line", value=False, elem_id=self.elem_id("checkbox_iterate"))
checkbox_iterate_batch = gr.Checkbox(label="Use same random seed for all lines", value=False, elem_id=self.elem_id("checkbox_iterate_batch"))
prompt_txt = gr.Textbox(label="List of prompt inputs", lines=1)
file = gr.File(label="Upload prompt inputs", type='bytes')
prompt_txt = gr.Textbox(label="List of prompt inputs", lines=1, elem_id=self.elem_id("prompt_txt"))
file = gr.File(label="Upload prompt inputs", type='bytes', elem_id=self.elem_id("file"))
file.change(fn=load_prompt_file, inputs=[file], outputs=[file, prompt_txt, prompt_txt])

10
scripts/sd_upscale.py
View File

@ -16,15 +16,17 @@ class Script(scripts.Script):
def show(self, is_img2img):
return is_img2img
def ui(self, is_img2img):
def ui(self, is_img2img):
info = gr.HTML("<p style=\"margin-bottom:0.75em\">Will upscale the image by the selected scale factor; use width and height sliders to set tile size</p>")
overlap = gr.Slider(minimum=0, maximum=256, step=16, label='Tile overlap', value=64)
scale_factor = gr.Slider(minimum=1, maximum=4, step=1, label='Scale Factor', value=2)
upscaler_index = gr.Radio(label='Upscaler', choices=[x.name for x in shared.sd_upscalers], value=shared.sd_upscalers[0].name, type="index")
overlap = gr.Slider(minimum=0, maximum=256, step=16, label='Tile overlap', value=64, elem_id=self.elem_id("overlap"))
scale_factor = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label='Scale Factor', value=2.0, elem_id=self.elem_id("scale_factor"))
upscaler_index = gr.Radio(label='Upscaler', choices=[x.name for x in shared.sd_upscalers], value=shared.sd_upscalers[0].name, type="index", elem_id=self.elem_id("upscaler_index"))
return [info, overlap, upscaler_index, scale_factor]
def run(self, p, _, overlap, upscaler_index, scale_factor):
if isinstance(upscaler_index, str):
upscaler_index = [x.name.lower() for x in shared.sd_upscalers].index(upscaler_index.lower())
processing.fix_seed(p)
upscaler = shared.sd_upscalers[upscaler_index]

48
scripts/xy_grid.py
View File

@ -10,7 +10,7 @@ import numpy as np
import modules.scripts as scripts
import gradio as gr
from modules import images, paths, sd_samplers
from modules import images, paths, sd_samplers, processing
from modules.hypernetworks import hypernetwork
from modules.processing import process_images, Processed, StableDiffusionProcessingTxt2Img
from modules.shared import opts, cmd_opts, state
@ -202,7 +202,7 @@ axis_options = [
AxisOption("Eta", float, apply_field("eta"), format_value_add_label, None),
AxisOption("Clip skip", int, apply_clip_skip, format_value_add_label, None),
AxisOption("Denoising", float, apply_field("denoising_strength"), format_value_add_label, None),
AxisOption("Upscale latent space for hires.", str, apply_upscale_latent_space, format_value_add_label, None),
AxisOption("Hires upscaler", str, apply_field("hr_upscaler"), format_value_add_label, None),
AxisOption("Cond. Image Mask Weight", float, apply_field("inpainting_mask_weight"), format_value_add_label, None),
AxisOption("VAE", str, apply_vae, format_value_add_label, None),
AxisOption("Styles", str, apply_styles, format_value_add_label, None),
@ -267,7 +267,6 @@ class SharedSettingsStackHelper(object):
self.CLIP_stop_at_last_layers = opts.CLIP_stop_at_last_layers
self.hypernetwork = opts.sd_hypernetwork
self.model = shared.sd_model
self.use_scale_latent_for_hires_fix = opts.use_scale_latent_for_hires_fix
self.vae = opts.sd_vae
def __exit__(self, exc_type, exc_value, tb):
@ -278,7 +277,6 @@ class SharedSettingsStackHelper(object):
hypernetwork.apply_strength()
opts.data["CLIP_stop_at_last_layers"] = self.CLIP_stop_at_last_layers
opts.data["use_scale_latent_for_hires_fix"] = self.use_scale_latent_for_hires_fix
re_range = re.compile(r"\s*([+-]?\s*\d+)\s*-\s*([+-]?\s*\d+)(?:\s*\(([+-]\d+)\s*\))?\s*")
@ -287,6 +285,7 @@ re_range_float = re.compile(r"\s*([+-]?\s*\d+(?:.\d*)?)\s*-\s*([+-]?\s*\d+(?:.\d
re_range_count = re.compile(r"\s*([+-]?\s*\d+)\s*-\s*([+-]?\s*\d+)(?:\s*\[(\d+)\s*\])?\s*")
re_range_count_float = re.compile(r"\s*([+-]?\s*\d+(?:.\d*)?)\s*-\s*([+-]?\s*\d+(?:.\d*)?)(?:\s*\[(\d+(?:.\d*)?)\s*\])?\s*")
class Script(scripts.Script):
def title(self):
return "X/Y plot"
@ -295,16 +294,16 @@ class Script(scripts.Script):
current_axis_options = [x for x in axis_options if type(x) == AxisOption or type(x) == AxisOptionImg2Img and is_img2img]
with gr.Row():
x_type = gr.Dropdown(label="X type", choices=[x.label for x in current_axis_options], value=current_axis_options[1].label, type="index", elem_id="x_type")
x_values = gr.Textbox(label="X values", lines=1)
x_type = gr.Dropdown(label="X type", choices=[x.label for x in current_axis_options], value=current_axis_options[1].label, type="index", elem_id=self.elem_id("x_type"))
x_values = gr.Textbox(label="X values", lines=1, elem_id=self.elem_id("x_values"))
with gr.Row():
y_type = gr.Dropdown(label="Y type", choices=[x.label for x in current_axis_options], value=current_axis_options[0].label, type="index", elem_id="y_type")
y_values = gr.Textbox(label="Y values", lines=1)
y_type = gr.Dropdown(label="Y type", choices=[x.label for x in current_axis_options], value=current_axis_options[0].label, type="index", elem_id=self.elem_id("y_type"))
y_values = gr.Textbox(label="Y values", lines=1, elem_id=self.elem_id("y_values"))
draw_legend = gr.Checkbox(label='Draw legend', value=True)
include_lone_images = gr.Checkbox(label='Include Separate Images', value=False)
no_fixed_seeds = gr.Checkbox(label='Keep -1 for seeds', value=False)
draw_legend = gr.Checkbox(label='Draw legend', value=True, elem_id=self.elem_id("draw_legend"))
include_lone_images = gr.Checkbox(label='Include Separate Images', value=False, elem_id=self.elem_id("include_lone_images"))
no_fixed_seeds = gr.Checkbox(label='Keep -1 for seeds', value=False, elem_id=self.elem_id("no_fixed_seeds"))
return [x_type, x_values, y_type, y_values, draw_legend, include_lone_images, no_fixed_seeds]
@ -383,7 +382,7 @@ class Script(scripts.Script):
ys = process_axis(y_opt, y_values)
def fix_axis_seeds(axis_opt, axis_list):
if axis_opt.label in ['Seed','Var. seed']:
if axis_opt.label in ['Seed', 'Var. seed']:
return [int(random.randrange(4294967294)) if val is None or val == '' or val == -1 else val for val in axis_list]
else:
return axis_list
@ -405,12 +404,33 @@ class Script(scripts.Script):
print(f"X/Y plot will create {len(xs) * len(ys) * p.n_iter} images on a {len(xs)}x{len(ys)} grid. (Total steps to process: {total_steps * p.n_iter})")
shared.total_tqdm.updateTotal(total_steps * p.n_iter)
grid_infotext = [None]
def cell(x, y):
pc = copy(p)
x_opt.apply(pc, x, xs)
y_opt.apply(pc, y, ys)
return process_images(pc)
res = process_images(pc)
if grid_infotext[0] is None:
pc.extra_generation_params = copy(pc.extra_generation_params)
if x_opt.label != 'Nothing':
pc.extra_generation_params["X Type"] = x_opt.label
pc.extra_generation_params["X Values"] = x_values
if x_opt.label in ["Seed", "Var. seed"] and not no_fixed_seeds:
pc.extra_generation_params["Fixed X Values"] = ", ".join([str(x) for x in xs])
if y_opt.label != 'Nothing':
pc.extra_generation_params["Y Type"] = y_opt.label
pc.extra_generation_params["Y Values"] = y_values
if y_opt.label in ["Seed", "Var. seed"] and not no_fixed_seeds:
pc.extra_generation_params["Fixed Y Values"] = ", ".join([str(y) for y in ys])
grid_infotext[0] = processing.create_infotext(pc, pc.all_prompts, pc.all_seeds, pc.all_subseeds)
return res
with SharedSettingsStackHelper():
processed = draw_xy_grid(
@ -425,6 +445,6 @@ class Script(scripts.Script):
)
if opts.grid_save:
images.save_image(processed.images[0], p.outpath_grids, "xy_grid", extension=opts.grid_format, prompt=p.prompt, seed=processed.seed, grid=True, p=p)
images.save_image(processed.images[0], p.outpath_grids, "xy_grid", info=grid_infotext[0], extension=opts.grid_format, prompt=p.prompt, seed=processed.seed, grid=True, p=p)
return processed

137
style.css
View File

@ -73,8 +73,9 @@
margin-right: auto;
}
#random_seed, #random_subseed, #reuse_seed, #reuse_subseed, #open_folder{
min-width: auto;
[id$=_random_seed], [id$=_random_subseed], [id$=_reuse_seed], [id$=_reuse_subseed], #open_folder{
min-width: 2.3em;
height: 2.5em;
flex-grow: 0;
padding-left: 0.25em;
padding-right: 0.25em;
@ -84,27 +85,28 @@
display: none;
}
#seed_row, #subseed_row{
[id$=_seed_row], [id$=_subseed_row]{
gap: 0.5rem;
padding: 0.6em;
}
#subseed_show_box{
[id$=_subseed_show_box]{
min-width: auto;
flex-grow: 0;
}
#subseed_show_box > div{
[id$=_subseed_show_box] > div{
border: 0;
height: 100%;
}
#subseed_show{
[id$=_subseed_show]{
min-width: auto;
flex-grow: 0;
padding: 0;
}
#subseed_show label{
[id$=_subseed_show] label{
height: 100%;
}
@ -206,24 +208,24 @@ button{
fieldset span.text-gray-500, .gr-block.gr-box span.text-gray-500, label.block span{
position: absolute;
top: -0.6em;
top: -0.7em;
line-height: 1.2em;
padding: 0 0.5em;
margin: 0;
padding: 0;
margin: 0 0.5em;
background-color: white;
border-top: 1px solid #eee;
border-left: 1px solid #eee;
border-right: 1px solid #eee;
box-shadow: 6px 0 6px 0px white, -6px 0 6px 0px white;
z-index: 300;
}
.dark fieldset span.text-gray-500, .dark .gr-block.gr-box span.text-gray-500, .dark label.block span{
background-color: rgb(31, 41, 55);
border-top: 1px solid rgb(55 65 81);
border-left: 1px solid rgb(55 65 81);
border-right: 1px solid rgb(55 65 81);
box-shadow: 6px 0 6px 0px rgb(31, 41, 55), -6px 0 6px 0px rgb(31, 41, 55);
}
#txt2img_column_batch, #img2img_column_batch{
min-width: min(13.5em, 100%) !important;
}
#settings fieldset span.text-gray-500, #settings .gr-block.gr-box span.text-gray-500, #settings label.block span{
@ -232,15 +234,38 @@ fieldset span.text-gray-500, .gr-block.gr-box span.text-gray-500, label.block s
margin-right: 8em;
}
.gr-panel div.flex-col div.justify-between label span{
margin: 0;
}
#settings .gr-panel div.flex-col div.justify-between div{
position: relative;
z-index: 200;
}
#settings{
display: block;
}
#settings > div{
border: none;
margin-left: 10em;
}
#settings > div.flex-wrap{
float: left;
display: block;
margin-left: 0;
width: 10em;
}
#settings > div.flex-wrap button{
display: block;
border: none;
text-align: left;
}
#settings_result{
height: 1.4em;
margin: 0 1.2em;
}
input[type="range"]{
margin: 0.5em 0 -0.3em 0;
}
@ -487,7 +512,7 @@ input[type="range"]{
border: none;
background: none;
flex: unset;
gap: 0.5em;
gap: 1em;
}
#quicksettings > div > div{
@ -496,6 +521,17 @@ input[type="range"]{
padding: 0;
}
#quicksettings > div > div > div > div > label > span {
position: relative;
margin-right: 9em;
margin-bottom: -1em;
}
#quicksettings > div > div > label > span {
position: relative;
margin-bottom: -1em;
}
canvas[key="mask"] {
z-index: 12 !important;
filter: invert();
@ -509,7 +545,7 @@ canvas[key="mask"] {
position: absolute;
right: 0.5em;
top: -0.6em;
z-index: 200;
z-index: 400;
width: 8em;
}
#quicksettings .gr-box > div > div > input.gr-text-input {
@ -530,7 +566,7 @@ img2maskimg, #img2maskimg > .h-60, #img2maskimg > .h-60 > div, #img2maskimg > .h
/* Extensions */
#tab_extensions table{
#tab_extensions table``{
border-collapse: collapse;
}
@ -556,6 +592,15 @@ img2maskimg, #img2maskimg > .h-60, #img2maskimg > .h-60 > div, #img2maskimg > .h
font-size: 95%;
}
#available_extensions .info{
margin: 0;
}
#available_extensions .date_added{
opacity: 0.85;
font-size: 90%;
}
#image_buttons_txt2img button, #image_buttons_img2img button, #image_buttons_extras button{
min-width: auto;
padding-left: 0.5em;
@ -582,6 +627,52 @@ img2maskimg, #img2maskimg > .h-60, #img2maskimg > .h-60 > div, #img2maskimg > .h
}
#img2img_settings > div.gr-form, #txt2img_settings > div.gr-form {
padding-top: 0.9em;
}
#img2img_settings div.gr-form .gr-form, #txt2img_settings div.gr-form .gr-form, #train_tabs div.gr-form .gr-form{
border: none;
padding-bottom: 0.5em;
}
footer {
display: none !important;
}
#footer{
text-align: center;
}
#footer div{
display: inline-block;
}
#footer .versions{
font-size: 85%;
opacity: 0.85;
}
#txtimg_hr_finalres{
min-height: 0 !important;
padding: .625rem .75rem;
margin-left: -0.75em
}
#txtimg_hr_finalres .resolution{
font-weight: bold;
}
#txt2img_checkboxes > div > div{
flex: 0;
white-space: nowrap;
min-width: auto;
}
.inactive{
opacity: 0.5;
}
/* The following handles localization for right-to-left (RTL) languages like Arabic.
The rtl media type will only be activated by the logic in javascript/localization.js.

0
test/advanced_features/__init__.py
View File

29
test/advanced_features/extras_test.py
View File

@ -1,29 +0,0 @@
import unittest
class TestExtrasWorking(unittest.TestCase):
def setUp(self):
self.url_img2img = "http://localhost:7860/sdapi/v1/extra-single-image"
self.simple_extras = {
"resize_mode": 0,
"show_extras_results": True,
"gfpgan_visibility": 0,
"codeformer_visibility": 0,
"codeformer_weight": 0,
"upscaling_resize": 2,
"upscaling_resize_w": 128,
"upscaling_resize_h": 128,
"upscaling_crop": True,
"upscaler_1": "None",
"upscaler_2": "None",
"extras_upscaler_2_visibility": 0,
"image": ""
}
class TestExtrasCorrectness(unittest.TestCase):
pass
if __name__ == "__main__":
unittest.main()

47
test/advanced_features/txt2img_test.py
View File

@ -1,47 +0,0 @@
import unittest
import requests
class TestTxt2ImgWorking(unittest.TestCase):
def setUp(self):
self.url_txt2img = "http://localhost:7860/sdapi/v1/txt2img"
self.simple_txt2img = {
"enable_hr": False,
"denoising_strength": 0,
"firstphase_width": 0,
"firstphase_height": 0,
"prompt": "example prompt",
"styles": [],
"seed": -1,
"subseed": -1,
"subseed_strength": 0,
"seed_resize_from_h": -1,
"seed_resize_from_w": -1,
"batch_size": 1,
"n_iter": 1,
"steps": 3,
"cfg_scale": 7,
"width": 64,
"height": 64,
"restore_faces": False,
"tiling": False,
"negative_prompt": "",
"eta": 0,
"s_churn": 0,
"s_tmax": 0,
"s_tmin": 0,
"s_noise": 1,
"sampler_index": "Euler a"
}
def test_txt2img_with_restore_faces_performed(self):
self.simple_txt2img["restore_faces"] = True
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
class TestTxt2ImgCorrectness(unittest.TestCase):
pass
if __name__ == "__main__":
unittest.main()

54
test/basic_features/extras_test.py Normal file
View File

@ -0,0 +1,54 @@
import unittest
import requests
from gradio.processing_utils import encode_pil_to_base64
from PIL import Image
class TestExtrasWorking(unittest.TestCase):
def setUp(self):
self.url_extras_single = "http://localhost:7860/sdapi/v1/extra-single-image"
self.extras_single = {
"resize_mode": 0,
"show_extras_results": True,
"gfpgan_visibility": 0,
"codeformer_visibility": 0,
"codeformer_weight": 0,
"upscaling_resize": 2,
"upscaling_resize_w": 128,
"upscaling_resize_h": 128,
"upscaling_crop": True,
"upscaler_1": "None",
"upscaler_2": "None",
"extras_upscaler_2_visibility": 0,
"image": encode_pil_to_base64(Image.open(r"test/test_files/img2img_basic.png"))
}
def test_simple_upscaling_performed(self):
self.extras_single["upscaler_1"] = "Lanczos"
self.assertEqual(requests.post(self.url_extras_single, json=self.extras_single).status_code, 200)
class TestPngInfoWorking(unittest.TestCase):
def setUp(self):
self.url_png_info = "http://localhost:7860/sdapi/v1/extra-single-image"
self.png_info = {
"image": encode_pil_to_base64(Image.open(r"test/test_files/img2img_basic.png"))
}
def test_png_info_performed(self):
self.assertEqual(requests.post(self.url_png_info, json=self.png_info).status_code, 200)
class TestInterrogateWorking(unittest.TestCase):
def setUp(self):
self.url_interrogate = "http://localhost:7860/sdapi/v1/extra-single-image"
self.interrogate = {
"image": encode_pil_to_base64(Image.open(r"test/test_files/img2img_basic.png")),
"model": "clip"
}
def test_interrogate_performed(self):
self.assertEqual(requests.post(self.url_interrogate, json=self.interrogate).status_code, 200)
if __name__ == "__main__":
unittest.main()

13
test/basic_features/img2img_test.py
View File

@ -16,7 +16,7 @@ class TestImg2ImgWorking(unittest.TestCase):
"inpainting_fill": 0,
"inpaint_full_res": False,
"inpaint_full_res_padding": 0,
"inpainting_mask_invert": 0,
"inpainting_mask_invert": False,
"prompt": "example prompt",
"styles": [],
"seed": -1,
@ -50,6 +50,17 @@ class TestImg2ImgWorking(unittest.TestCase):
self.simple_img2img["mask"] = encode_pil_to_base64(Image.open(r"test/test_files/mask_basic.png"))
self.assertEqual(requests.post(self.url_img2img, json=self.simple_img2img).status_code, 200)
def test_inpainting_with_inverted_masked_performed(self):
self.simple_img2img["mask"] = encode_pil_to_base64(Image.open(r"test/test_files/mask_basic.png"))
self.simple_img2img["inpainting_mask_invert"] = True
self.assertEqual(requests.post(self.url_img2img, json=self.simple_img2img).status_code, 200)
def test_img2img_sd_upscale_performed(self):
self.simple_img2img["script_name"] = "sd upscale"
self.simple_img2img["script_args"] = ["", 8, "Lanczos", 2.0]
self.assertEqual(requests.post(self.url_img2img, json=self.simple_img2img).status_code, 200)
if __name__ == "__main__":
unittest.main()

12
test/basic_features/txt2img_test.py
View File

@ -41,6 +41,10 @@ class TestTxt2ImgWorking(unittest.TestCase):
self.simple_txt2img["negative_prompt"] = "example negative prompt"
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
def test_txt2img_with_complex_prompt_performed(self):
self.simple_txt2img["prompt"] = "((emphasis)), (emphasis1:1.1), [to:1], [from::2], [from:to:0.3], [alt|alt1]"
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
def test_txt2img_not_square_image_performed(self):
self.simple_txt2img["height"] = 128
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
@ -53,6 +57,10 @@ class TestTxt2ImgWorking(unittest.TestCase):
self.simple_txt2img["tiling"] = True
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
def test_txt2img_with_restore_faces_performed(self):
self.simple_txt2img["restore_faces"] = True
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
def test_txt2img_with_vanilla_sampler_performed(self):
self.simple_txt2img["sampler_index"] = "PLMS"
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
@ -63,6 +71,10 @@ class TestTxt2ImgWorking(unittest.TestCase):
self.simple_txt2img["n_iter"] = 2
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
def test_txt2img_batch_performed(self):
self.simple_txt2img["batch_size"] = 2
self.assertEqual(requests.post(self.url_txt2img, json=self.simple_txt2img).status_code, 200)
if __name__ == "__main__":
unittest.main()

17
test/basic_features/utils_test.py
View File

@ -14,10 +14,25 @@ class UtilsTests(unittest.TestCase):
self.url_prompt_styles = "http://localhost:7860/sdapi/v1/prompt-styles"
self.url_artist_categories = "http://localhost:7860/sdapi/v1/artist-categories"
self.url_artists = "http://localhost:7860/sdapi/v1/artists"
self.url_embeddings = "http://localhost:7860/sdapi/v1/embeddings"
def test_options_get(self):
self.assertEqual(requests.get(self.url_options).status_code, 200)
def test_options_write(self):
response = requests.get(self.url_options)
self.assertEqual(response.status_code, 200)
pre_value = response.json()["send_seed"]
self.assertEqual(requests.post(self.url_options, json={"send_seed":not pre_value}).status_code, 200)
response = requests.get(self.url_options)
self.assertEqual(response.status_code, 200)
self.assertEqual(response.json()["send_seed"], not pre_value)
requests.post(self.url_options, json={"send_seed": pre_value})
def test_cmd_flags(self):
self.assertEqual(requests.get(self.url_cmd_flags).status_code, 200)
@ -48,6 +63,8 @@ class UtilsTests(unittest.TestCase):
def test_artists(self):
self.assertEqual(requests.get(self.url_artists).status_code, 200)
def test_embeddings(self):
self.assertEqual(requests.get(self.url_artists).status_code, 200)
if __name__ == "__main__":
unittest.main()

2
test/server_poll.py
View File

@ -15,7 +15,7 @@ def run_tests(proc, test_dir):
break
if proc.poll() is None:
if test_dir is None:
test_dir = ""
test_dir = "test"
suite = unittest.TestLoader().discover(test_dir, pattern="*_test.py", top_level_dir="test")
result = unittest.TextTestRunner(verbosity=2).run(suite)
return len(result.failures) + len(result.errors)

BIN
txt2img_Screenshot.png
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 329 KiB

44
webui.py
View File

@ -1,17 +1,23 @@
import os
import sys
import threading
import time
import importlib
import signal
import threading
import re
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from fastapi.middleware.gzip import GZipMiddleware
from modules import import_hook
from modules import import_hook, errors
from modules.call_queue import wrap_queued_call, queue_lock, wrap_gradio_gpu_call
from modules.paths import script_path
import torch
# Truncate version number of nightly/local build of PyTorch to not cause exceptions with CodeFormer or Safetensors
if ".dev" in torch.__version__ or "+git" in torch.__version__:
torch.__version__ = re.search(r'[\d.]+[\d]', torch.__version__).group(0)
from modules import shared, devices, sd_samplers, upscaler, extensions, localization, ui_tempdir
import modules.codeformer_model as codeformer
import modules.extras
@ -27,6 +33,7 @@ import modules.sd_models
import modules.sd_vae
import modules.txt2img
import modules.script_callbacks
import modules.textual_inversion.textual_inversion
import modules.ui
from modules import modelloader
@ -55,12 +62,22 @@ def initialize():
gfpgan.setup_model(cmd_opts.gfpgan_models_path)
shared.face_restorers.append(modules.face_restoration.FaceRestoration())
modelloader.list_builtin_upscalers()
modules.scripts.load_scripts()
modelloader.load_upscalers()
modules.sd_vae.refresh_vae_list()
modules.sd_models.load_model()
modules.textual_inversion.textual_inversion.list_textual_inversion_templates()
try:
modules.sd_models.load_model()
except Exception as e:
errors.display(e, "loading stable diffusion model")
print("", file=sys.stderr)
print("Stable diffusion model failed to load, exiting", file=sys.stderr)
exit(1)
shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: modules.sd_models.reload_model_weights()))
shared.opts.onchange("sd_vae", wrap_queued_call(lambda: modules.sd_vae.reload_vae_weights()), call=False)
shared.opts.onchange("sd_vae_as_default", wrap_queued_call(lambda: modules.sd_vae.reload_vae_weights()), call=False)
@ -91,11 +108,11 @@ def initialize():
def setup_cors(app):
if cmd_opts.cors_allow_origins and cmd_opts.cors_allow_origins_regex:
app.add_middleware(CORSMiddleware, allow_origins=cmd_opts.cors_allow_origins.split(','), allow_origin_regex=cmd_opts.cors_allow_origins_regex, allow_methods=['*'])
app.add_middleware(CORSMiddleware, allow_origins=cmd_opts.cors_allow_origins.split(','), allow_origin_regex=cmd_opts.cors_allow_origins_regex, allow_methods=['*'], allow_credentials=True, allow_headers=['*'])
elif cmd_opts.cors_allow_origins:
app.add_middleware(CORSMiddleware, allow_origins=cmd_opts.cors_allow_origins.split(','), allow_methods=['*'])
app.add_middleware(CORSMiddleware, allow_origins=cmd_opts.cors_allow_origins.split(','), allow_methods=['*'], allow_credentials=True, allow_headers=['*'])
elif cmd_opts.cors_allow_origins_regex:
app.add_middleware(CORSMiddleware, allow_origin_regex=cmd_opts.cors_allow_origins_regex, allow_methods=['*'])
app.add_middleware(CORSMiddleware, allow_origin_regex=cmd_opts.cors_allow_origins_regex, allow_methods=['*'], allow_credentials=True, allow_headers=['*'])
def create_api(app):
@ -169,23 +186,24 @@ def webui():
modules.script_callbacks.app_started_callback(shared.demo, app)
wait_on_server(shared.demo)
print('Restarting UI...')
sd_samplers.set_samplers()
print('Reloading extensions')
modules.script_callbacks.script_unloaded_callback()
extensions.list_extensions()
localization.list_localizations(cmd_opts.localizations_dir)
print('Reloading custom scripts')
modelloader.forbid_loaded_nonbuiltin_upscalers()
modules.scripts.reload_scripts()
modules.script_callbacks.model_loaded_callback(shared.sd_model)
modelloader.load_upscalers()
print('Reloading modules: modules.ui')
importlib.reload(modules.ui)
print('Refreshing Model List')
for module in [module for name, module in sys.modules.items() if name.startswith("modules.ui")]:
importlib.reload(module)
modules.sd_models.list_models()
print('Restarting Gradio')
if __name__ == "__main__":

4
webui.sh
View File

@ -160,10 +160,10 @@ then
printf "\n%s\n" "${delimiter}"
printf "Accelerating launch.py..."
printf "\n%s\n" "${delimiter}"
accelerate launch --num_cpu_threads_per_process=6 "${LAUNCH_SCRIPT}" "$@"
exec accelerate launch --num_cpu_threads_per_process=6 "${LAUNCH_SCRIPT}" "$@"
else
printf "\n%s\n" "${delimiter}"
printf "Launching launch.py..."
printf "\n%s\n" "${delimiter}"
"${python_cmd}" "${LAUNCH_SCRIPT}" "$@"
exec "${python_cmd}" "${LAUNCH_SCRIPT}" "$@"
fi