From d61a5aa4f623f6630670241aca8fc5c2a6381769 Mon Sep 17 00:00:00 2001 From: acncagua Date: Fri, 6 Jan 2023 10:58:22 +0900 Subject: Add files via upload --- modules/ui.py | 2 ++ 1 file changed, 2 insertions(+) (limited to 'modules/ui.py') diff --git a/modules/ui.py b/modules/ui.py index 81d96c5b..030f0685 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -550,6 +550,8 @@ Requested path was: {f} os.startfile(path) elif platform.system() == "Darwin": sp.Popen(["open", path]) + elif "microsoft-standard-WSL2" in platform.uname().release: + sp.Popen(["wsl-open", path]) else: sp.Popen(["xdg-open", path]) -- cgit v1.2.3 From 3246a2d6b898da6a98fe9df4dc67944635a41bd3 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Fri, 6 Jan 2023 16:03:43 +0300 Subject: remove restriction for saving dropdowns to ui-config.json --- modules/scripts.py | 1 - modules/ui.py | 10 ++-------- 2 files changed, 2 insertions(+), 9 deletions(-) (limited to 'modules/ui.py') diff --git a/modules/scripts.py b/modules/scripts.py index 0c44f191..35164093 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -290,7 +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: diff --git a/modules/ui.py b/modules/ui.py index 030f0685..b79d24ee 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -435,11 +435,9 @@ def create_toprow(is_img2img): with gr.Row(): with gr.Column(scale=1, elem_id="style_pos_col"): prompt_style = gr.Dropdown(label="Style 1", elem_id=f"{id_part}_style_index", choices=[k for k, v in shared.prompt_styles.styles.items()], value=next(iter(shared.prompt_styles.styles.keys()))) - prompt_style.save_to_config = True with gr.Column(scale=1, elem_id="style_neg_col"): prompt_style2 = gr.Dropdown(label="Style 2", elem_id=f"{id_part}_style2_index", choices=[k for k, v in shared.prompt_styles.styles.items()], value=next(iter(shared.prompt_styles.styles.keys()))) - prompt_style2.save_to_config = True return prompt, prompt_style, negative_prompt, prompt_style2, submit, button_interrogate, button_deepbooru, prompt_style_apply, save_style, paste, token_counter, token_button @@ -638,7 +636,6 @@ def create_sampler_and_steps_selection(choices, tabname): if opts.samplers_in_dropdown: with FormRow(elem_id=f"sampler_selection_{tabname}"): sampler_index = gr.Dropdown(label='Sampling method', elem_id=f"{tabname}_sampling", choices=[x.name for x in choices], value=choices[0].name, type="index") - sampler_index.save_to_config = True steps = gr.Slider(minimum=1, maximum=150, step=1, elem_id=f"{tabname}_steps", label="Sampling steps", value=20) else: with FormGroup(elem_id=f"sampler_selection_{tabname}"): @@ -1794,7 +1791,7 @@ def create_ui(): if init_field is not None: init_field(saved_value) - if type(x) in [gr.Slider, gr.Radio, gr.Checkbox, gr.Textbox, gr.Number] and x.visible: + if type(x) in [gr.Slider, gr.Radio, gr.Checkbox, gr.Textbox, gr.Number, gr.Dropdown] and x.visible: apply_field(x, 'visible') if type(x) == gr.Slider: @@ -1815,11 +1812,8 @@ def create_ui(): if type(x) == gr.Number: apply_field(x, 'value') - # Since there are many dropdowns that shouldn't be saved, - # we only mark dropdowns that should be saved. - if type(x) == gr.Dropdown and getattr(x, 'save_to_config', False): + if type(x) == gr.Dropdown: apply_field(x, 'value', lambda val: val in x.choices, getattr(x, 'init_field', None)) - apply_field(x, 'visible') visit(txt2img_interface, loadsave, "txt2img") visit(img2img_interface, loadsave, "img2img") -- cgit v1.2.3 From 3992ecbe6e46a465062508c677964534e7397f72 Mon Sep 17 00:00:00 2001 From: Mitchell Boot <47387831+Mitchell1711@users.noreply.github.com> Date: Fri, 6 Jan 2023 18:02:46 +0100 Subject: Added UI elements Added a new row to hires fix that shows the new resolution after scaling --- modules/ui.py | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'modules/ui.py') diff --git a/modules/ui.py b/modules/ui.py index b79d24ee..20f7d2a2 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -255,6 +255,12 @@ def add_style(name: str, prompt: str, negative_prompt: str): return [gr.Dropdown.update(visible=True, choices=list(shared.prompt_styles.styles)) for _ in range(4)] +def calc_resolution_hires(x, y, scale): + #final res can only be a multiple of 8 + scaled_x = int(x * scale // 8) * 8 + scaled_y = int(y * scale // 8) * 8 + + return "

Upscaled Resolution: "+str(scaled_x)+"x"+str(scaled_y)+"

" def apply_styles(prompt, prompt_neg, style1_name, style2_name): prompt = shared.prompt_styles.apply_styles_to_prompt(prompt, [style1_name, style2_name]) @@ -718,6 +724,12 @@ def create_ui(): hr_scale = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label="Upscale by", value=2.0, elem_id="txt2img_hr_scale") hr_resize_x = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize width to", value=0, elem_id="txt2img_hr_resize_x") hr_resize_y = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize height to", value=0, elem_id="txt2img_hr_resize_y") + + with FormRow(elem_id="txt2img_hires_fix_row3"): + hr_final_resolution = gr.HTML(value=calc_resolution_hires(width.value, height.value, hr_scale.value), elem_id="txtimg_hr_finalres") + hr_scale.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) + width.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) + height.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) elif category == "batch": if not opts.dimensions_and_batch_together: -- cgit v1.2.3 From 991368c8d54404d8e13d4c6e76a0f32644e65ad4 Mon Sep 17 00:00:00 2001 From: Mitchell Boot <47387831+Mitchell1711@users.noreply.github.com> Date: Fri, 6 Jan 2023 18:24:29 +0100 Subject: remove camelcase --- modules/ui.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'modules/ui.py') diff --git a/modules/ui.py b/modules/ui.py index 20f7d2a2..6fc8b7d7 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -260,7 +260,7 @@ def calc_resolution_hires(x, y, scale): scaled_x = int(x * scale // 8) * 8 scaled_y = int(y * scale // 8) * 8 - return "

Upscaled Resolution: "+str(scaled_x)+"x"+str(scaled_y)+"

" + return "

Upscaled resolution: "+str(scaled_x)+"x"+str(scaled_y)+"

" def apply_styles(prompt, prompt_neg, style1_name, style2_name): prompt = shared.prompt_styles.apply_styles_to_prompt(prompt, [style1_name, style2_name]) -- cgit v1.2.3 From 79e39fae6110c20a3ee6255e2841c877f65e8cbd Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Sat, 7 Jan 2023 01:45:28 +0300 Subject: CLIP hijack rework --- modules/sd_hijack.py | 6 +- modules/sd_hijack_clip.py | 348 ++++++++++++------------- modules/sd_hijack_clip_old.py | 81 ++++++ modules/textual_inversion/textual_inversion.py | 1 - modules/ui.py | 2 +- 5 files changed, 256 insertions(+), 182 deletions(-) create mode 100644 modules/sd_hijack_clip_old.py (limited to 'modules/ui.py') diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index fa2cd4bb..71cc145a 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -150,10 +150,10 @@ 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): diff --git a/modules/sd_hijack_clip.py b/modules/sd_hijack_clip.py index ca92b142..ac3020d7 100644 --- a/modules/sd_hijack_clip.py +++ b/modules/sd_hijack_clip.py @@ -1,12 +1,28 @@ import math +from collections import namedtuple import torch from modules import prompt_parser, devices 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']) +"""This is a marker showing that textual inversion embedding's vectors have to placed at offset in the prompt chunk""" class FrozenCLIPEmbedderWithCustomWordsBase(torch.nn.Module): @@ -14,17 +30,49 @@ class FrozenCLIPEmbedderWithCustomWordsBase(torch.nn.Module): super().__init__() self.wrapped = wrapped self.hijack = 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 + """ + 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 +80,152 @@ 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 + + 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 - embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i) + 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) + + # 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 is there is a comma nearby, the text after 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 + + reloc_tokens = chunk.tokens[break_location:] + reloc_mults = chunk.multipliers[break_location:] - remade_tokens = remade_tokens[:last_comma] - length = len(remade_tokens) + chunk.tokens = chunk.tokens[:break_location] + chunk.multipliers = chunk.multipliers[:break_location] - 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 + 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 - - token_count = len(remade_tokens) - prompt_target_length = get_target_prompt_token_count(token_count) - tokens_to_add = prompt_target_length - len(remade_tokens) - - remade_tokens = remade_tokens + [self.id_end] * tokens_to_add - multipliers = multipliers + [1.0] * tokens_to_add - - return remade_tokens, fixes, multipliers, token_count - - def process_text(self, texts): - used_custom_terms = [] - remade_batch_tokens = [] - hijack_comments = [] - hijack_fixes = [] + chunk.tokens.append(token) + chunk.multipliers.append(weight) + position += 1 + continue + + emb_len = int(embedding.vec.shape[0]) + if len(chunk.tokens) + emb_len > self.chunk_length: + next_chunk() + + chunk.fixes.append(PromptChunkFix(len(chunk.tokens), embedding)) + + chunk.tokens += [0] * emb_len + chunk.multipliers += [weight] * emb_len + position += embedding_length_in_tokens + + if len(chunk.tokens) > 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. + """ + 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 - - 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 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(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 + batch_chunks, token_count = self.process_texts(texts) - return z + used_embeddings = {} + chunk_count = max([len(x) for x in batch_chunks]) - 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] + zs = [] + for i in range(chunk_count): + batch_chunk = [chunks[i] if i < len(chunks) else self.empty_chunk() for chunks in batch_chunks] + + 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] + for fixes in self.hijack.fixes: + for position, embedding in fixes: + used_embeddings[embedding.name] = embedding + + z = self.process_tokens(tokens, multipliers) + zs.append(z) + + 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}") + + return torch.hstack(zs) + + def process_tokens(self, remade_batch_tokens, 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) @@ -239,8 +234,7 @@ 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) new_mean = z.mean() diff --git a/modules/sd_hijack_clip_old.py b/modules/sd_hijack_clip_old.py new file mode 100644 index 00000000..6d9fbbe6 --- /dev/null +++ b/modules/sd_hijack_clip_old.py @@ -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) diff --git a/modules/textual_inversion/textual_inversion.py b/modules/textual_inversion/textual_inversion.py index f9f5e8cd..45882ed6 100644 --- a/modules/textual_inversion/textual_inversion.py +++ b/modules/textual_inversion/textual_inversion.py @@ -79,7 +79,6 @@ class EmbeddingDatabase: 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] diff --git a/modules/ui.py b/modules/ui.py index b79d24ee..5d2f5bad 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -368,7 +368,7 @@ def update_token_counter(text, steps): flat_prompts = reduce(lambda list1, list2: list1+list2, prompt_schedules) prompts = [prompt_text for step, prompt_text in flat_prompts] - tokens, token_count, max_length = max([model_hijack.tokenize(prompt) for prompt in prompts], key=lambda args: args[1]) + token_count, max_length = max([model_hijack.get_prompt_lengths(prompt) for prompt in prompts], key=lambda args: args[0]) style_class = ' class="red"' if (token_count > max_length) else "" return f"{token_count}/{max_length}" -- cgit v1.2.3 From f94cfc563bbedd923d5e95563a5e8d93c8516ac3 Mon Sep 17 00:00:00 2001 From: Mitchell Boot <47387831+Mitchell1711@users.noreply.github.com> Date: Sat, 7 Jan 2023 01:15:22 +0100 Subject: Changed HTML to textbox instead Using HTML caused an issue where the row would expand for a frame when changing the sliders because of the loading animation. This solution also doesn't use any additional HTML padding --- modules/ui.py | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'modules/ui.py') diff --git a/modules/ui.py b/modules/ui.py index 6fc8b7d7..6ea1b5d7 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -260,7 +260,7 @@ def calc_resolution_hires(x, y, scale): scaled_x = int(x * scale // 8) * 8 scaled_y = int(y * scale // 8) * 8 - return "

Upscaled resolution: "+str(scaled_x)+"x"+str(scaled_y)+"

" + return str(scaled_x)+"x"+str(scaled_y) def apply_styles(prompt, prompt_neg, style1_name, style2_name): prompt = shared.prompt_styles.apply_styles_to_prompt(prompt, [style1_name, style2_name]) @@ -726,7 +726,10 @@ def create_ui(): hr_resize_y = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize height to", value=0, elem_id="txt2img_hr_resize_y") with FormRow(elem_id="txt2img_hires_fix_row3"): - hr_final_resolution = gr.HTML(value=calc_resolution_hires(width.value, height.value, hr_scale.value), elem_id="txtimg_hr_finalres") + hr_final_resolution = gr.Textbox(value=calc_resolution_hires(width.value, height.value, hr_scale.value), + elem_id="txtimg_hr_finalres", + label="Upscaled resolution", + interactive=False) hr_scale.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) width.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) height.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) -- cgit v1.2.3 From de9738044571877450d1038e18f1ecce93d24af3 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Sat, 7 Jan 2023 08:53:53 +0300 Subject: this breaks on default config because width, height, hr_scale are None at that point. --- modules/ui.py | 14 ++++++-------- 1 file changed, 6 insertions(+), 8 deletions(-) (limited to 'modules/ui.py') diff --git a/modules/ui.py b/modules/ui.py index f946382d..a18b9007 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -725,14 +725,8 @@ def create_ui(): hr_resize_x = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize width to", value=0, elem_id="txt2img_hr_resize_x") hr_resize_y = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize height to", value=0, elem_id="txt2img_hr_resize_y") - with FormRow(elem_id="txt2img_hires_fix_row3"): - hr_final_resolution = gr.Textbox(value=calc_resolution_hires(width.value, height.value, hr_scale.value), - elem_id="txtimg_hr_finalres", - label="Upscaled resolution", - interactive=False) - hr_scale.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) - width.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) - height.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) + with FormRow(elem_id="txt2img_hires_fix_row3"): + hr_final_resolution = gr.Textbox(value="", elem_id="txtimg_hr_finalres", label="Upscaled resolution", interactive=False) elif category == "batch": if not opts.dimensions_and_batch_together: @@ -744,6 +738,10 @@ def create_ui(): with FormGroup(elem_id="txt2img_script_container"): custom_inputs = modules.scripts.scripts_txt2img.setup_ui() + hr_scale.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) + width.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) + height.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) + txt2img_gallery, generation_info, html_info, html_log = create_output_panel("txt2img", opts.outdir_txt2img_samples) parameters_copypaste.bind_buttons({"txt2img": txt2img_paste}, None, txt2img_prompt) -- cgit v1.2.3 From 1a5b86ad65fd738eadea1ad72f4abad3a4aabf17 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Sat, 7 Jan 2023 09:56:37 +0300 Subject: rework hires fix preview for #6437: movie it to where it takes less place, make it actually account for all relevant sliders and calculate dimensions correctly --- modules/processing.py | 1 - modules/ui.py | 40 +++++++++++++++++++++++++++------------- modules/ui_components.py | 8 ++++++++ style.css | 17 +++++++++++++++++ 4 files changed, 52 insertions(+), 14 deletions(-) (limited to 'modules/ui.py') diff --git a/modules/processing.py b/modules/processing.py index a408d622..82157bc9 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -711,7 +711,6 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): self.truncate_x = 0 self.truncate_y = 0 - def init(self, all_prompts, all_seeds, all_subseeds): if self.enable_hr: if self.hr_resize_x == 0 and self.hr_resize_y == 0: diff --git a/modules/ui.py b/modules/ui.py index a18b9007..6c765262 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -20,7 +20,7 @@ from PIL import Image, PngImagePlugin from modules.call_queue import wrap_gradio_gpu_call, wrap_queued_call, wrap_gradio_call from modules import sd_hijack, sd_models, localization, script_callbacks, ui_extensions, deepbooru -from modules.ui_components import FormRow, FormGroup, ToolButton +from modules.ui_components import FormRow, FormGroup, ToolButton, FormHTML from modules.paths import script_path from modules.shared import opts, cmd_opts, restricted_opts @@ -255,12 +255,20 @@ def add_style(name: str, prompt: str, negative_prompt: str): return [gr.Dropdown.update(visible=True, choices=list(shared.prompt_styles.styles)) for _ in range(4)] -def calc_resolution_hires(x, y, scale): - #final res can only be a multiple of 8 - scaled_x = int(x * scale // 8) * 8 - scaled_y = int(y * scale // 8) * 8 - - return str(scaled_x)+"x"+str(scaled_y) + +def calc_resolution_hires(enable, width, height, hr_scale, hr_resize_x, hr_resize_y): + from modules import processing, devices + + if not enable: + return "" + + p = processing.StableDiffusionProcessingTxt2Img(width=width, height=height, enable_hr=True, hr_scale=hr_scale, hr_resize_x=hr_resize_x, hr_resize_y=hr_resize_y) + + with devices.autocast(): + p.init([""], [0], [0]) + + return f"resize to: {p.hr_upscale_to_x}x{p.hr_upscale_to_y}" + def apply_styles(prompt, prompt_neg, style1_name, style2_name): prompt = shared.prompt_styles.apply_styles_to_prompt(prompt, [style1_name, style2_name]) @@ -712,6 +720,7 @@ def create_ui(): restore_faces = gr.Checkbox(label='Restore faces', value=False, visible=len(shared.face_restorers) > 1, elem_id="txt2img_restore_faces") tiling = gr.Checkbox(label='Tiling', value=False, elem_id="txt2img_tiling") enable_hr = gr.Checkbox(label='Hires. fix', value=False, elem_id="txt2img_enable_hr") + hr_final_resolution = FormHTML(value="", elem_id="txtimg_hr_finalres", label="Upscaled resolution", interactive=False) elif category == "hires_fix": with FormGroup(visible=False, elem_id="txt2img_hires_fix") as hr_options: @@ -724,9 +733,6 @@ def create_ui(): hr_scale = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label="Upscale by", value=2.0, elem_id="txt2img_hr_scale") hr_resize_x = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize width to", value=0, elem_id="txt2img_hr_resize_x") hr_resize_y = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize height to", value=0, elem_id="txt2img_hr_resize_y") - - with FormRow(elem_id="txt2img_hires_fix_row3"): - hr_final_resolution = gr.Textbox(value="", elem_id="txtimg_hr_finalres", label="Upscaled resolution", interactive=False) elif category == "batch": if not opts.dimensions_and_batch_together: @@ -738,9 +744,16 @@ def create_ui(): with FormGroup(elem_id="txt2img_script_container"): custom_inputs = modules.scripts.scripts_txt2img.setup_ui() - hr_scale.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) - width.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) - height.change(fn=calc_resolution_hires, inputs=[width, height, hr_scale], outputs=hr_final_resolution, show_progress=False) + hr_resolution_preview_inputs = [enable_hr, width, height, hr_scale, hr_resize_x, hr_resize_y] + hr_resolution_preview_args = dict( + fn=calc_resolution_hires, + inputs=hr_resolution_preview_inputs, + outputs=[hr_final_resolution], + show_progress=False + ) + + for input in hr_resolution_preview_inputs: + input.change(**hr_resolution_preview_args) txt2img_gallery, generation_info, html_info, html_log = create_output_panel("txt2img", opts.outdir_txt2img_samples) parameters_copypaste.bind_buttons({"txt2img": txt2img_paste}, None, txt2img_prompt) @@ -803,6 +816,7 @@ def create_ui(): fn=lambda x: gr_show(x), inputs=[enable_hr], outputs=[hr_options], + show_progress = False, ) txt2img_paste_fields = [ diff --git a/modules/ui_components.py b/modules/ui_components.py index 91eb0e3d..cac001dc 100644 --- a/modules/ui_components.py +++ b/modules/ui_components.py @@ -23,3 +23,11 @@ class FormGroup(gr.Group, gr.components.FormComponent): 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" + diff --git a/style.css b/style.css index f1b23b53..76721756 100644 --- a/style.css +++ b/style.css @@ -642,6 +642,23 @@ footer { 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; +} + /* 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. If you change anything above, you need to make sure it is RTL compliant by just running -- cgit v1.2.3 From fdfce4711076c2ebac1089bac8169d043eb7978f Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Sat, 7 Jan 2023 13:29:47 +0300 Subject: add "from" resolution for hires fix to be less confusing. --- modules/ui.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'modules/ui.py') diff --git a/modules/ui.py b/modules/ui.py index 6c765262..99483130 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -267,7 +267,7 @@ def calc_resolution_hires(enable, width, height, hr_scale, hr_resize_x, hr_resiz with devices.autocast(): p.init([""], [0], [0]) - return f"resize to: {p.hr_upscale_to_x}x{p.hr_upscale_to_y}" + return f"resize: from {width}x{height} to {p.hr_upscale_to_x}x{p.hr_upscale_to_y}" def apply_styles(prompt, prompt_neg, style1_name, style2_name): -- cgit v1.2.3