stable-diffusion-utils/utils/renamer/preprocess.py

143 lines
4.3 KiB
Python
Executable File

# Credits to https://gist.github.com/nnuudev/56ed3242023c8582a32e3130ef59730b / https://boards.4chan.org/trash/thread/51463059#p51472156
import os
import re
import json
import time
import shutil
import urllib.request
config = {
'input': './in/', # files to process
'output': './out/', # files to copy files to
'tags': './tags.csv', # csv of tags associated with the yiffy model (replace for other flavor of booru's taglist associated with the model you're training against)
'cache': './cache.json', # JSON file of cached tags, will speed up processing if re-running
'rateLimit': 500, # time to wait between requests, in milliseconds, e621 imposes a rate limit of 2 requests per second
'filenameLimit': 192, # maximum characters to put in the filename, necessary to abide by filesystem limitations, and to "limit" token count for the prompt parser
'filter': True,
# fill it with tags of whatever you don't want to make it into the filename
# for starters, you can also add "anthro", "male", "female", as they're very common tags
'filters': [
"anthro",
"fur",
"male",
"female",
"genitals",
"video games",
"animal genitalia",
],
'filtersRegex': [
r"clothing$",
],
'onlyIncludeModelArtists': True, # if True, only include the artist's tag if in the model's taglist, if false, add all artists
# i've noticed some artists that weren't included in the taglist, but is available in LAION's (vanilla SD)
'reverseTags': False, # inverts sorting, prioritizing tags with little representation in the model
}
with open(config['tags'], 'rb') as f:
csv = f.read().decode('utf-8').split("\n")
config['tags'] = {}
for i in csv:
k, v = i.split(',')
config['tags'][k] = int(v)
cache = {}
try:
with open(config['cache'], 'rb') as f:
cache = json.loads(f.read().decode('utf-8'))
except:
pass
def parse():
global config, cache
files = []
for file in os.listdir(config['input']):
files.append(file)
for i in range(len(files)):
file = files[i]
md5 = re.match(r"^([a-f0-9]{32})", file)
if not md5:
continue
md5 = md5.group(1)
print(f"[{(100.0 * i / len(files)):3.0f}%]: {md5}")
rateLimit = False
if not md5 in cache:
rateLimit = True
with urllib.request.urlopen(urllib.request.Request(f"https://e621.net/posts.json?tags=md5:{md5}",
headers = {
'user-agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/87.0.4280.88 Safari/537.36'
}
)) as r:
j = json.loads(r.read())
cache[md5] = j["posts"][0] if j["posts"] else {}
json_meta = cache[md5]
if not json_meta:
continue
tags = []
artist = ""
content = {
"s": "safe content",
"q": "questionable content",
"e": "explict content",
}.get(json_meta["rating"], "")
for cat in json_meta["tags"]:
if cat == "artist":
tag = "by " + " and ".join(json_meta["tags"]["artist"])
if config['onlyIncludeModelArtists'] and not tag in config['tags']:
continue
artist = tag
else:
for tag in json_meta["tags"][cat]:
tag = tag.replace("_", " ")
if tag not in config['tags']:
continue
if "/" in tag or ":" in tag:
continue # illegal filename character
if config['filter']:
should = False
if tag in config['filters']:
continue # was break in the original script, fixed ;)
for filter in config['filtersRegex']:
if re.search(filter, tag):
should = True
continue # was break in the original script, fixed ;)
if should:
continue
tags.append(tag)
tags.sort(key=lambda x: -config['tags'][x], reverse=config['reverseTags'])
if artist:
tags.insert(0, artist)
if content:
tags.insert(0, content)
jointmp = ""
filtered = []
for i in tags:
if len(jointmp + " " + i) > config['filenameLimit']:
break
jointmp += " " + i
filtered.append(i)
joined = " ".join(filtered)
shutil.copy(os.path.join(config['input'], file), os.path.join(config['output'], file.replace(md5, joined)))
if rateLimit and config['rateLimit']:
time.sleep(config['rateLimit'] / 1000.0)
# NOOOOOOOO YOU'RE WASTING SPACE BY PRETTIFYING
with open(config['cache'], 'wb') as f:
f.write(json.dumps(cache, indent='\t').encode('utf-8'))
if __name__ == "__main__":
parse()