stable-diffusion-webui/modules/prompt_parser.py

import re
from collections import namedtuple

import lark

# a prompt like this: "fantasy landscape with a [mountain:lake:0.25] and [an oak:a christmas tree:0.75][ in foreground::0.6][ in background:0.25] [shoddy:masterful:0.5]"
# will be represented with prompt_schedule like this (assuming steps=100):
# [25, 'fantasy landscape with a mountain and an oak in foreground shoddy']
# [50, 'fantasy landscape with a lake and an oak in foreground in background shoddy']
# [60, 'fantasy landscape with a lake and an oak in foreground in background masterful']
# [75, 'fantasy landscape with a lake and an oak in background masterful']
# [100, 'fantasy landscape with a lake and a christmas tree in background masterful']

schedule_parser = lark.Lark(r"""
!start: (prompt | /[][():]/+)*
prompt: (emphasized | scheduled | plain | WHITESPACE)*
!emphasized: "(" prompt ")"
        | "(" prompt ":" prompt ")"
        | "[" prompt "]"
scheduled: "[" [prompt ":"] prompt ":" [WHITESPACE] NUMBER "]"
WHITESPACE: /\s+/
plain: /([^\\\[\]():]|\\.)+/
%import common.SIGNED_NUMBER -> NUMBER
""")

def get_learned_conditioning_prompt_schedules(prompts, steps):
    """
    >>> g = lambda p: get_learned_conditioning_prompt_schedules([p], 10)[0]
    >>> g("test")
    [[10, 'test']]
    >>> g("a [b:3]")
    [[3, 'a '], [10, 'a b']]
    >>> g("a [b: 3]")
    [[3, 'a '], [10, 'a b']]
    >>> g("a [[[b]]:2]")
    [[2, 'a '], [10, 'a [[b]]']]
    >>> g("[(a:2):3]")
    [[3, ''], [10, '(a:2)']]
    >>> g("a [b : c : 1] d")
    [[1, 'a b  d'], [10, 'a  c  d']]
    >>> g("a[b:[c:d:2]:1]e")
    [[1, 'abe'], [2, 'ace'], [10, 'ade']]
    >>> g("a [unbalanced")
    [[10, 'a [unbalanced']]
    >>> g("a [b:.5] c")
    [[5, 'a  c'], [10, 'a b c']]
    >>> g("a [{b|d{:.5] c")  # not handling this right now
    [[5, 'a  c'], [10, 'a {b|d{ c']]
    >>> g("((a][:b:c [d:3]")
    [[3, '((a][:b:c '], [10, '((a][:b:c d']]
    """

    def collect_steps(steps, tree):
        l = [steps]
        class CollectSteps(lark.Visitor):
            def scheduled(self, tree):
                tree.children[-1] = float(tree.children[-1])
                if tree.children[-1] < 1:
                    tree.children[-1] *= steps
                tree.children[-1] = min(steps, int(tree.children[-1]))
                l.append(tree.children[-1])
        CollectSteps().visit(tree)
        return sorted(set(l))

    def at_step(step, tree):
        class AtStep(lark.Transformer):
            def scheduled(self, args):
                before, after, _, when = args
                yield before or () if step <= when else after
            def start(self, args):
                def flatten(x):
                    if type(x) == str:
                        yield x
                    else:
                        for gen in x:
                            yield from flatten(gen)
                return ''.join(flatten(args))
            def plain(self, args):
                yield args[0].value
            def __default__(self, data, children, meta):
                for child in children:
                    yield from child
        return AtStep().transform(tree)

    def get_schedule(prompt):
        try:
            tree = schedule_parser.parse(prompt)
        except lark.exceptions.LarkError as e:
            if 0:
                import traceback
                traceback.print_exc()
            return [[steps, prompt]]
        return [[t, at_step(t, tree)] for t in collect_steps(steps, tree)]

    promptdict = {prompt: get_schedule(prompt) for prompt in set(prompts)}
    return [promptdict[prompt] for prompt in prompts]


ScheduledPromptConditioning = namedtuple("ScheduledPromptConditioning", ["end_at_step", "cond"])
ScheduledPromptBatch = namedtuple("ScheduledPromptBatch", ["shape", "schedules"])


def get_learned_conditioning(model, prompts, steps):
    res = []

    prompt_schedules = get_learned_conditioning_prompt_schedules(prompts, steps)
    cache = {}

    for prompt, prompt_schedule in zip(prompts, prompt_schedules):

        cached = cache.get(prompt, None)
        if cached is not None:
            res.append(cached)
            continue

        texts = [x[1] for x in prompt_schedule]
        conds = model.get_learned_conditioning(texts)

        cond_schedule = []
        for i, (end_at_step, text) in enumerate(prompt_schedule):
            cond_schedule.append(ScheduledPromptConditioning(end_at_step, conds[i]))

        cache[prompt] = cond_schedule
        res.append(cond_schedule)

    return ScheduledPromptBatch((len(prompts),) + res[0][0].cond.shape, res)


def reconstruct_cond_batch(c: ScheduledPromptBatch, current_step):
    param = c.schedules[0][0].cond
    res = torch.zeros(c.shape, device=param.device, dtype=param.dtype)
    for i, cond_schedule in enumerate(c.schedules):
        target_index = 0
        for current, (end_at, cond) in enumerate(cond_schedule):
            if current_step <= end_at:
                target_index = current
                break
        res[i] = cond_schedule[target_index].cond

    return res


re_attention = re.compile(r"""
\\\(|
\\\)|
\\\[|
\\]|
\\\\|
\\|
\(|
\[|
:([+-]?[.\d]+)\)|
\)|
]|
[^\\()\[\]:]+|
:
""", re.X)


def parse_prompt_attention(text):
    """
    Parses a string with attention tokens and returns a list of pairs: text and its assoicated weight.
    Accepted tokens are:
      (abc) - increases attention to abc by a multiplier of 1.1
      (abc:3.12) - increases attention to abc by a multiplier of 3.12
      [abc] - decreases attention to abc by a multiplier of 1.1
      \( - literal character '('
      \[ - literal character '['
      \) - literal character ')'
      \] - literal character ']'
      \\ - literal character '\'
      anything else - just text

    >>> parse_prompt_attention('normal text')
    [['normal text', 1.0]]
    >>> parse_prompt_attention('an (important) word')
    [['an ', 1.0], ['important', 1.1], [' word', 1.0]]
    >>> parse_prompt_attention('(unbalanced')
    [['unbalanced', 1.1]]
    >>> parse_prompt_attention('\(literal\]')
    [['(literal]', 1.0]]
    >>> parse_prompt_attention('(unnecessary)(parens)')
    [['unnecessaryparens', 1.1]]
    >>> parse_prompt_attention('a (((house:1.3)) [on] a (hill:0.5), sun, (((sky))).')
    [['a ', 1.0],
     ['house', 1.5730000000000004],
     [' ', 1.1],
     ['on', 1.0],
     [' a ', 1.1],
     ['hill', 0.55],
     [', sun, ', 1.1],
     ['sky', 1.4641000000000006],
     ['.', 1.1]]
    """

    res = []
    round_brackets = []
    square_brackets = []

    round_bracket_multiplier = 1.1
    square_bracket_multiplier = 1 / 1.1

    def multiply_range(start_position, multiplier):
        for p in range(start_position, len(res)):
            res[p][1] *= multiplier

    for m in re_attention.finditer(text):
        text = m.group(0)
        weight = m.group(1)

        if text.startswith('\\'):
            res.append([text[1:], 1.0])
        elif text == '(':
            round_brackets.append(len(res))
        elif text == '[':
            square_brackets.append(len(res))
        elif weight is not None and len(round_brackets) > 0:
            multiply_range(round_brackets.pop(), float(weight))
        elif text == ')' and len(round_brackets) > 0:
            multiply_range(round_brackets.pop(), round_bracket_multiplier)
        elif text == ']' and len(square_brackets) > 0:
            multiply_range(square_brackets.pop(), square_bracket_multiplier)
        else:
            res.append([text, 1.0])

    for pos in round_brackets:
        multiply_range(pos, round_bracket_multiplier)

    for pos in square_brackets:
        multiply_range(pos, square_bracket_multiplier)

    if len(res) == 0:
        res = [["", 1.0]]

    # merge runs of identical weights
    i = 0
    while i + 1 < len(res):
        if res[i][1] == res[i + 1][1]:
            res[i][0] += res[i + 1][0]
            res.pop(i + 1)
        else:
            i += 1

    return res

if __name__ == "__main__":
    import doctest
    doctest.testmod(optionflags=doctest.NORMALIZE_WHITESPACE)
else:
    import torch  # doctest faster