-
Notifications
You must be signed in to change notification settings - Fork 2
/
sdworker.py
364 lines (293 loc) · 15 KB
/
sdworker.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
#!/usr/bin/env python3
import json
import io
import logging
import os
import pathlib
import random
import shutil
import sys
from typing import NamedTuple
import time
import traceback
from diffusers import StableDiffusionPipeline, DiffusionPipeline, EulerAncestralDiscreteScheduler
from PIL.Image import Image
from PIL.PngImagePlugin import PngInfo
import torch
from safetensors.torch import load_file
REPLACERS_FILEPATH="config/replacers.json"
OLD_REPLACER_FILEPATH="replacers.json"
REPLACER_SAMPLE_FILEPATH="replacers.json.sample"
SpecialTag = NamedTuple('SpecialTag', words=list[str], join_word=str, min=int, max=int, max_occurences=int)
logging.basicConfig(stream=sys.stdout, level=logging.DEBUG)
class DeguDiffusionWorker():
def __init__(self, sd_token:str, output_folder:str="", save_to_disk:bool=True, model_name:str="CompVis/stable-diffusion-v1-4", mode:str="fp32", local_only:bool=False, sd_cache_dir:str="", torch_device="cuda", additional_model=""):
# Test
logger = logging.getLogger('DeguDiffusionWorker')
logger.info('Initializing StableDiffusion')
self.logger = logger
self.model_name = model_name
self.torch_device = torch_device
self.save_to_disk = save_to_disk
if save_to_disk:
if not output_folder:
raise ValueError(f"No output directory provided")
if not os.path.isdir(output_folder):
raise ValueError(f"The provided images output path doesn't point to a directory :\n{output_folder}")
pipeline_kwargs = dict()
pipeline_kwargs["use_safetensors"] = True
pipeline_kwargs["add_watermarker"] = False
if sd_token:
pipeline_kwargs["use_auth_token"] = sd_token
if sd_cache_dir:
if not os.path.isdir(sd_cache_dir):
raise ValueError(f"The provided StableDiffusion cache path doesn't point to a directory :\n{sd_cache_dir}")
pipeline_kwargs["cache_dir"] = sd_cache_dir
if local_only:
pipeline_kwargs["local_files_only"] = True
if mode == "fp16":
pipeline_kwargs["variant"] = "fp16"
pipeline_kwargs["torch_dtype"] = torch.float16
pipeline_kwargs["torch_dtype"] = torch.float16
#scheduler = DDIMScheduler.from_pretrained(self.model_name, subfolder="scheduler", **pipeline_kwargs)
#scheduler = DPMSolverMultistepScheduler.from_pretrained(self.model_name, subfolder="scheduler")
#pipeline_kwargs["scheduler"] = scheduler
if not self.model_name.startswith("./"):
pipe = DiffusionPipeline.from_pretrained(self.model_name, **pipeline_kwargs)
else:
pipe = StableDiffusionPipeline.from_single_file(self.model_name, **pipeline_kwargs)
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
#pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
#pipe = StableDiffusionPipeline.from_pretrained(pathlib.Path("./stablediffusion_cache/nai"), **pipeline_kwargs)
pipe.to(self.torch_device)
#pipe.enable_attention_slicing()
logger.debug(str(pipe))
# Worker specific values
self.output_folder:pathlib.Path = pathlib.Path(output_folder) if output_folder else None
self.pipe = pipe
self.results = {}
self.replacers:dict = self.load_replacers(
replacers_filepath = REPLACERS_FILEPATH,
sample_filepath = REPLACER_SAMPLE_FILEPATH,
old_replacers_filepath = OLD_REPLACER_FILEPATH)
logger.info(f"Using model {model_name}")
logger.info("StableDiffusion ready to go")
def generate_image(
self,
prompt: str = "",
n_inferences: int = 50,
guidance_scale: float = 7.5,
deterministic = True,
width:int = 512,
height:int = 512):
report = {}
generator = None
seed = 'Unknown'
if deterministic:
if type(deterministic) is int:
seed = deterministic
else:
seed = torch.Generator(self.torch_device).seed()
generator = torch.Generator(self.torch_device).manual_seed(seed)
original_prompt = prompt
prompt = self.replace_special_tags(prompt, self.replacers)
report["actual_prompt"] = prompt if original_prompt != prompt else ""
negative_prompt = "lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name"
metadata = PngInfo()
metadata.add_itxt("AI_Prompt", str(prompt), lang="utf8", tkey="AI_Prompt")
metadata.add_text("AI_Torch_Seed", str(seed))
metadata.add_text("AI_StableDiffusion_Guidance_Scale", str(guidance_scale))
metadata.add_text("AI_StableDiffusion_Inferences", str(n_inferences))
metadata.add_text("AI_StableDiffusion_Model_Name", str(self.model_name))
#metadata.add_text("AI_Diffusers_Version", str(self.pipe._diffusers_version))
metadata.add_text("AI_Metadata_Type", "Voyage")
metadata.add_text("AI_Metadata_Voyage_Version", "0")
metadata.add_text("AI_Generator", str(self.model_name))
metadata.add_text("AI_Torch_Generator", str(self.torch_device))
metadata.add_text("AI_Custom_Deterministic", str(deterministic))
metadata.add_itxt("AI_Prompt_Negative", str(negative_prompt), lang="utf8", tkey="AI_Prompt_Negative")
metadata.add_text("AI_StableDiffusion_Pipe", str(self.pipe))
#with torch.autocast(self.torch_device):
result = self.pipe(
prompt,
negative_prompt = negative_prompt,
width=width,
height=height,
guidance_scale=guidance_scale,
generator=generator,
num_inference_steps=n_inferences)
nsfw_flag = False # result["nsfw_content_detected"][0]
report["seed"] = seed
report["nsfw"] = nsfw_flag
report["filepath"] = ""
report["content_as"] = "file" if self.save_to_disk else "data"
if not nsfw_flag:
image:Image = result.images[0]
filename = f"{int(time.time())}_SEED_{seed}.png"
if self.output_folder:
filepath = self.output_folder / filename
else:
filepath = filename
report["filepath"] = filepath
if self.save_to_disk:
image.save(filepath, pnginfo=metadata)
else:
image_data = io.BytesIO()
image.save(image_data, format='PNG', pnginfo=metadata)
image_data.seek(0)
report["image_data"] = image_data
return report
def load_replacers(
self,
replacers_filepath = "config/replacers.json",
sample_filepath = "replacers.json.sample",
old_replacers_filepath = "replacers.json") -> dict:
replacers:dict = dict()
if not os.path.exists(replacers_filepath):
self.logger.debug(f"[load_replacers] {replacers_filepath} does not exist.")
if os.path.exists(old_replacers_filepath):
self.logger.debug(f"Copying previous {old_replacers_filepath} to {replacers_filepath}")
shutil.copy2(old_replacers_filepath, replacers_filepath)
else:
if not os.path.exists(sample_filepath):
self.logger.error("No sample available... Something is very wrong with your installation...")
return replacers
self.logger.debug(f"Copying the sample {sample_filepath}")
shutil.copy2(sample_filepath, replacers_filepath)
if not os.path.isfile(replacers_filepath):
self.logger.error(f"[load_replacers] {replacers_filepath} is not a file ?? Giving up about replacers")
return replacers
with open(replacers_filepath, 'r', encoding='utf-8') as f:
try:
json_content = json.load(f)
except Exception as e:
traceback.print_exception(e)
self.logger.error(f"[load_replacers] An error happened when trying to parse the JSON file")
return replacers
if not 'replacements' in json_content:
self.logger.error(f"[load_replacers] No 'replacements' section in root of the JSON file {replacers_filepath}")
return replacers
replacements = json_content['replacements']
replacements_field_type = type(replacements)
if not replacements_field_type == dict:
self.logger.error(f"[load_replacers] replacements must be an OBJECT (dict).\nCurrently it is a {replacements_field_type}")
return replacers
required_fields = {
"words": list,
"join_word": str,
"min": int,
"max": int,
"max_occurences": int
}
required_field_keys = required_fields.keys()
for item_name in json_content['replacements']:
item = replacements[item_name]
if type(item) != dict:
self.logger.warning(f"Invalid type for {item}. Skipping")
continue
# Yet another obnoxious Python syntax
item_keys = item.keys()
if not (item.keys() >= required_field_keys):
self.logger.warning(f"Missing keys in {item}.\nKeys required : {str(required_field_keys)}\nGot : {str(item_keys)}")
continue
invalid_fields = []
for required_field in required_field_keys:
if type(item[required_field]) != required_fields[required_field]:
invalid_fields.append(required_field)
if invalid_fields:
for invalid_field in invalid_fields:
self.logger.warning(f"{invalid_fields} MUST be a {required_fields[invalid_field]}. Currently : {type(item[required_field])}")
continue
replacer = SpecialTag(
words=item["words"],
join_word=item["join_word"],
min=item["min"],
max=item["max"],
max_occurences=item["max_occurences"])
replacers[item_name] = replacer
return replacers
def random_from_tag(self, replacer:SpecialTag) -> list[str]:
names = replacer.words
n_names = random.randint(
min(replacer.min, len(names)),
min(replacer.max, len(names)))
return random.sample(names, n_names)
def replace_special_tags(self, prompt, tags:list[SpecialTag]):
for tag_name in tags:
tag:SpecialTag = tags[tag_name]
if tag == None:
self.logger.debug("Tag %s has no value ???" % (tag_name))
continue
if tag_name not in prompt:
continue
occurences = 0
max_occurences = tag.max_occurences
while tag_name in prompt:
if occurences >= max_occurences:
prompt = prompt.replace(tag_name, "")
break
names_list = self.random_from_tag(tag)
prompt = prompt.replace(tag_name, tag.join_word.join(names_list), 1)
occurences += 1
return prompt
if __name__ == "__main__":
import dotenv
from myylibs.helpers import Helpers
dotenv.load_dotenv()
logger = logging.getLogger('StableDiffusion standalone test')
# FIXME Factorize this with degu_diffusion into a specific python file.
# Basically, make a configuration object...
IMAGES_OUTPUT_DIRECTORY = os.environ.get('IMAGES_OUTPUT_DIRECTORY', 'generated')
IMAGES_WIDTH = Helpers.env_var_to_int('IMAGES_WIDTH', 512)
IMAGES_HEIGHT = Helpers.env_var_to_int('IMAGES_HEIGHT', 512)
STABLEDIFFUSION_LOCAL_ONLY = False if os.environ.get('STABLEDIFFUSION_LOCAL_ONLY', 'False').lower() != 'true' else True
HUGGINGFACES_TOKEN = os.environ.get('HUGGINGFACES_TOKEN', '')
STABLEDIFFUSION_MODEL_NAME = os.environ.get('STABLEDIFFUSION_MODEL_NAME', 'CompVis/stable-diffusion-v1-4')
TORCH_DEVICE = os.environ.get('TORCH_DEVICE', 'cuda')
STABLEDIFFUSION_CACHE_DIR = os.environ.get('STABLEDIFFUSION_CACHE_DIR', '')
SAFETENSORS_ADDITIONAL_MODEL = os.environ.get('SAFETENSORS_ADDITIONAL_MODEL', '')
if STABLEDIFFUSION_CACHE_DIR and (not os.path.exists(STABLEDIFFUSION_CACHE_DIR)):
pathlib.Path(STABLEDIFFUSION_CACHE_DIR).mkdir(parents = True)
if (not HUGGINGFACES_TOKEN) and (not STABLEDIFFUSION_LOCAL_ONLY):
logger.fatal(
"At least, either :\n"+
"* Set the HUGGINGFACES_TOKEN environment variable.\n"+
"* Set the STABLEDIFFUSION_LOCAL_ONLY environment variable to true\n"+
"You can also set both, in which case STABLEDIFFUSION_LOCAL_ONLY will take precedence when set to true")
exit(1)
if not os.path.exists(IMAGES_OUTPUT_DIRECTORY):
pathlib.Path(IMAGES_OUTPUT_DIRECTORY).mkdir(parents = True)
diffuser = DeguDiffusionWorker(
model_name = STABLEDIFFUSION_MODEL_NAME,
sd_token = os.environ.get('HUGGINGFACES_TOKEN', ''),
output_folder = IMAGES_OUTPUT_DIRECTORY,
mode = os.environ.get('STABLEDIFFUSION_MODE', 'fp32'),
sd_cache_dir = os.environ.get('STABLEDIFFUSION_CACHE_DIR', ''),
local_only = STABLEDIFFUSION_LOCAL_ONLY,
torch_device = TORCH_DEVICE,
additional_model = SAFETENSORS_ADDITIONAL_MODEL)
logger.info("Standalone Stable Diffusion test")
DEFAULT_IMAGES_PER_JOB = Helpers.env_var_to_int('DEFAULT_IMAGES_PER_JOB', 8)
# This is not a formatted string, don't add a f near the quotes
DEFAULT_PROMPT = os.environ.get('DEFAULT_PROMPT', 'Degu enjoys its morning coffee by {random_artists}, {random_tags}')
DEFAULT_SEED = os.environ.get('DEFAULT_SEED', '')
DEFAULT_INFERENCES_STEPS = Helpers.env_var_to_int('DEFAULT_INFERENCES_STEPS', 60)
DEFAULT_GUIDANCE_SCALE = Helpers.env_var_to_float('DEFAULT_GUIDANCE_SCALE', 7.5)
SEED_MINUS_ONE_IS_RANDOM = True if os.environ.get('SEED_MINUS_ONE_IS_RANDOM', 'True').lower() != "false" else False
seed_value = None
if DEFAULT_SEED:
try:
seed_value = int(DEFAULT_SEED)
except ValueError:
pass
if seed_value == -1 and SEED_MINUS_ONE_IS_RANDOM:
seed_value = None
for _ in range(0, DEFAULT_IMAGES_PER_JOB):
diffuser.generate_image(
prompt = DEFAULT_PROMPT,
n_inferences = DEFAULT_INFERENCES_STEPS,
guidance_scale = DEFAULT_GUIDANCE_SCALE,
deterministic = seed_value if seed_value else True,
width = IMAGES_WIDTH,
height = IMAGES_HEIGHT)
logger.info(f"Test finished. Check the output in {IMAGES_OUTPUT_DIRECTORY}")