-
Notifications
You must be signed in to change notification settings - Fork 67
/
Copy pathpix2pix.py
executable file
·238 lines (209 loc) · 8.73 KB
/
pix2pix.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
# https://github.com/Woolverine94/biniou
# pix2pix.py
import gradio as gr
import os
import PIL
import torch
from diffusers import StableDiffusionInstructPix2PixPipeline, StableDiffusionXLInstructPix2PixPipeline
from compel import Compel, ReturnedEmbeddingsType
import random
from ressources.common import *
from ressources.gfpgan import *
import tomesd
device_label_pix2pix, model_arch = detect_device()
device_pix2pix = torch.device(device_label_pix2pix)
# Gestion des modèles -> pas concerné (safetensors refusé)
model_path_pix2pix = "./models/pix2pix/"
model_path_safety_checker = "./models/Stable_Diffusion/"
os.makedirs(model_path_pix2pix, exist_ok=True)
model_list_pix2pix = []
for filename in os.listdir(model_path_pix2pix):
f = os.path.join(model_path_pix2pix, filename)
if os.path.isfile(f) and (filename.endswith('.ckpt') or filename.endswith('.safetensors')):
model_list_pix2pix.append(f)
model_list_pix2pix_builtin = [
"timbrooks/instruct-pix2pix",
"instruction-tuning-sd/low-level-img-proc",
"instruction-tuning-sd/cartoonizer",
"diffusers/sdxl-instructpix2pix-768",
]
for k in range(len(model_list_pix2pix_builtin)):
model_list_pix2pix.append(model_list_pix2pix_builtin[k])
# Bouton Cancel
stop_pix2pix = False
def initiate_stop_pix2pix() :
global stop_pix2pix
stop_pix2pix = True
def check_pix2pix(pipe, step_index, timestep, callback_kwargs) :
global stop_pix2pix
if stop_pix2pix == False :
return callback_kwargs
elif stop_pix2pix == True :
print(">>>[Instruct pix2pix 🖌️ ]: generation canceled by user")
stop_pix2pix = False
try:
del ressources.pix2pix.pipe_pix2pix
except NameError as e:
raise Exception("Interrupting ...")
return
def check_pix2pix_xl(step, timestep, latents) :
global stop_pix2pix
if stop_pix2pix == False :
return
elif stop_pix2pix == True :
print(">>>[Instruct pix2pix 🖌️ ]: generation canceled by user")
stop_pix2pix = False
try:
del ressources.pix2pix.pipe_pix2pix
except NameError as e:
raise Exception("Interrupting ...")
return
@metrics_decoration
def image_pix2pix(
modelid_pix2pix,
sampler_pix2pix,
img_pix2pix,
prompt_pix2pix,
negative_prompt_pix2pix,
num_images_per_prompt_pix2pix,
num_prompt_pix2pix,
guidance_scale_pix2pix,
image_guidance_scale_pix2pix,
num_inference_step_pix2pix,
height_pix2pix,
width_pix2pix,
seed_pix2pix,
use_gfpgan_pix2pix,
nsfw_filter,
tkme_pix2pix,
progress_pix2pix=gr.Progress(track_tqdm=True)
):
print(">>>[Instruct pix2pix 🖌️ ]: starting module")
nsfw_filter_final, feat_ex = safety_checker_sd(model_path_safety_checker, device_pix2pix, nsfw_filter)
if is_sdxl(modelid_pix2pix):
is_xl_pix2pix: bool = True
else :
is_xl_pix2pix: bool = False
if is_xl_pix2pix:
pipe_pix2pix= StableDiffusionXLInstructPix2PixPipeline.from_pretrained(
modelid_pix2pix,
cache_dir=model_path_pix2pix,
torch_dtype=model_arch,
use_safetensors=True,
safety_checker=nsfw_filter_final,
feature_extractor=feat_ex,
resume_download=True,
local_files_only=True if offline_test() else None
)
else:
pipe_pix2pix= StableDiffusionInstructPix2PixPipeline.from_pretrained(
modelid_pix2pix,
cache_dir=model_path_pix2pix,
torch_dtype=model_arch,
use_safetensors=True if (modelid_pix2pix == "timbrooks/instruct-pix2pix") else False,
safety_checker=nsfw_filter_final,
feature_extractor=feat_ex,
resume_download=True,
local_files_only=True if offline_test() else None
)
pipe_pix2pix = schedulerer(pipe_pix2pix, sampler_pix2pix)
pipe_pix2pix.enable_attention_slicing("max")
tomesd.apply_patch(pipe_pix2pix, ratio=tkme_pix2pix)
if device_label_pix2pix == "cuda" :
pipe_pix2pix.enable_sequential_cpu_offload()
else :
pipe_pix2pix = pipe_pix2pix.to(device_pix2pix)
if seed_pix2pix == 0:
random_seed = torch.randint(0, 10000000000, (1,))
generator = torch.manual_seed(random_seed)
else:
generator = torch.manual_seed(seed_pix2pix)
if (is_xl_pix2pix == True):
dim_size = correct_size(width_pix2pix, height_pix2pix, 768)
else:
dim_size = correct_size(width_pix2pix, height_pix2pix, 512)
image_input = PIL.Image.open(img_pix2pix)
image_input = image_input.convert("RGB")
image_input = image_input.resize((dim_size[0], dim_size[1]))
width_pix2pix = dim_size[0]
height_pix2pix = dim_size[1]
prompt_pix2pix = str(prompt_pix2pix)
negative_prompt_pix2pix = str(negative_prompt_pix2pix)
if prompt_pix2pix == "None":
prompt_pix2pix = ""
if negative_prompt_pix2pix == "None":
negative_prompt_pix2pix = ""
if (is_xl_pix2pix == True):
compel = Compel(
tokenizer=[pipe_pix2pix.tokenizer, pipe_pix2pix.tokenizer_2],
text_encoder=[pipe_pix2pix.text_encoder, pipe_pix2pix.text_encoder_2],
returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
requires_pooled=[False, True],
device=device_pix2pix,
)
conditioning, pooled = compel(prompt_pix2pix)
neg_conditioning, neg_pooled = compel(negative_prompt_pix2pix)
[conditioning, neg_conditioning] = compel.pad_conditioning_tensors_to_same_length([conditioning, neg_conditioning])
else :
compel = Compel(tokenizer=pipe_pix2pix.tokenizer, text_encoder=pipe_pix2pix.text_encoder, truncate_long_prompts=False, device=device_pix2pix)
conditioning = compel.build_conditioning_tensor(prompt_pix2pix)
neg_conditioning = compel.build_conditioning_tensor(negative_prompt_pix2pix)
[conditioning, neg_conditioning] = compel.pad_conditioning_tensors_to_same_length([conditioning, neg_conditioning])
final_image = []
for i in range (num_prompt_pix2pix):
if (is_xl_pix2pix == True):
image = pipe_pix2pix(
image=image_input,
prompt_embeds=conditioning,
pooled_prompt_embeds=pooled,
negative_prompt_embeds=neg_conditioning,
negative_pooled_prompt_embeds=neg_pooled,
width=width_pix2pix,
height=height_pix2pix,
num_images_per_prompt=num_images_per_prompt_pix2pix,
guidance_scale=guidance_scale_pix2pix,
image_guidance_scale=image_guidance_scale_pix2pix,
num_inference_steps=num_inference_step_pix2pix,
generator = generator,
callback=check_pix2pix_xl,
).images
else:
image = pipe_pix2pix(
image=image_input,
prompt_embeds=conditioning,
negative_prompt_embeds=neg_conditioning,
width=width_pix2pix,
height=height_pix2pix,
num_images_per_prompt=num_images_per_prompt_pix2pix,
guidance_scale=guidance_scale_pix2pix,
image_guidance_scale=image_guidance_scale_pix2pix,
num_inference_steps=num_inference_step_pix2pix,
generator = generator,
callback_on_step_end=check_pix2pix,
callback_on_step_end_tensor_inputs=['latents'],
).images
for j in range(len(image)):
savename = name_image()
if use_gfpgan_pix2pix == True :
image[j] = image_gfpgan_mini(image[j])
image[j].save(savename)
final_image.append(savename)
print(f">>>[Instruct pix2pix 🖌️ ]: generated {num_prompt_pix2pix} batch(es) of {num_images_per_prompt_pix2pix}")
reporting_pix2pix = f">>>[Instruct pix2pix 🖌️ ]: "+\
f"Settings : Model={modelid_pix2pix} | "+\
f"Sampler={sampler_pix2pix} | "+\
f"Steps={num_inference_step_pix2pix} | "+\
f"CFG scale={guidance_scale_pix2pix} | "+\
f"Img CFG scale={image_guidance_scale_pix2pix} | "+\
f"Size={width_pix2pix}x{height_pix2pix} | "+\
f"GFPGAN={use_gfpgan_pix2pix} | "+\
f"Token merging={tkme_pix2pix} | "+\
f"nsfw_filter={bool(int(nsfw_filter))} | "+\
f"Prompt={prompt_pix2pix} | "+\
f"Negative prompt={negative_prompt_pix2pix}"
print(reporting_pix2pix)
exif_writer_png(reporting_pix2pix, final_image)
del nsfw_filter_final, feat_ex, pipe_pix2pix, generator, image_input, image
clean_ram()
print(f">>>[Instruct pix2pix 🖌️ ]: leaving module")
return final_image, final_image