Spaces:
Runtime error
Runtime error
File size: 19,638 Bytes
b888bcf 6a4b741 d3774b8 3d5a08b 6a4b741 ad569d5 5715833 6db905d f5d25ef 6a4b741 b888bcf f424501 2f833d2 b6a8c7c f5d25ef f5f53dc 15183c0 d3774b8 1f087be f5f53dc b888bcf c4cd17d 2f833d2 c4cd17d 6db905d 09de898 2f833d2 b6a8c7c c4cd17d b888bcf 6a4b741 d06267b ad569d5 2f833d2 ad569d5 2f833d2 b888bcf b6a8c7c f5d25ef 6130cf9 560d75d f5d25ef 3bef9b2 f5d25ef 71c1a49 f5d25ef 71c1a49 f5d25ef 71c1a49 f5d25ef 71c1a49 f5d25ef 71c1a49 bade8d8 f424501 6a4b741 6b7c1b1 b6a8c7c d06267b c4cd17d d06267b c4cd17d 09de898 5715833 d06267b 91f39f9 5715833 b6a8c7c 5715833 24d15b9 d06267b 5715833 6ba990e d06267b 74395e4 6ba990e 74395e4 8fe2fce 6ceab9c 8fe2fce be2828d c4cd17d be2828d 6f329ae 8ca8d03 10151ae 645b9bf 4c36274 8ca8d03 3d5a08b 8ca8d03 3d5a08b 3f3a00c 3d5a08b 8ca8d03 3d5a08b 3f3a00c 8ca8d03 dc9311b 8ca8d03 dc9311b 8ca8d03 10151ae 8ca8d03 10151ae 645b9bf 8ca8d03 1f087be 8ca8d03 3f3a00c dc9311b c24886c 3f3a00c dc9311b 645b9bf dc9311b d06267b 8ca8d03 523ae5e d06267b c24886c 8ca8d03 f5d25ef 8ca8d03 c4cd17d be2828d 8ca8d03 2f833d2 0cc4495 be2828d dc9311b 10151ae 0cc4495 1f087be b888bcf 2f833d2 6de6264 2f833d2 6de6264 b6a8c7c b888bcf 5715833 2f833d2 5715833 77c1441 554faea 5715833 d51f6a9 8f38828 eb191cf 8f38828 eb191cf 6de6264 eb191cf ce1bdbb eb191cf 16fac59 eb191cf a7ae1ff 8f38828 90a23cf d4fb1c8 90a23cf 2c0f2f0 6397acd 6a4b741 32b0f65 26a0ac1 050c639 26a0ac1 6de6264 b888bcf 2f833d2 d06267b b888bcf 2f833d2 b888bcf d06267b 560d75d c4cd17d 5715833 d06267b 5715833 b888bcf 8fe2fce c4cd17d 5715833 d06267b 5715833 b888bcf b6a8c7c 6de6264 8fe2fce b6a8c7c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 |
import gradio as gr
import torch
torch.jit.script = lambda f: f
import timm
import time
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
from share_btn import community_icon_html, loading_icon_html, share_js
from cog_sdxl_dataset_and_utils import TokenEmbeddingsHandler
import lora
import copy
import json
import gc
import random
from urllib.parse import quote
import gdown
import os
import diffusers
from diffusers.utils import load_image
from diffusers.models import ControlNetModel
from diffusers import AutoencoderKL, DPMSolverMultistepScheduler
import cv2
import torch
import numpy as np
from PIL import Image
from insightface.app import FaceAnalysis
from pipeline_stable_diffusion_xl_instantid_img2img import StableDiffusionXLInstantIDImg2ImgPipeline, draw_kps
from controlnet_aux import ZoeDetector
from compel import Compel, ReturnedEmbeddingsType
import spaces
#from gradio_imageslider import ImageSlider
with open("sdxl_loras.json", "r") as file:
data = json.load(file)
sdxl_loras_raw = [
{
"image": item["image"],
"title": item["title"],
"repo": item["repo"],
"trigger_word": item["trigger_word"],
"weights": item["weights"],
"is_compatible": item["is_compatible"],
"is_pivotal": item.get("is_pivotal", False),
"text_embedding_weights": item.get("text_embedding_weights", None),
"likes": item.get("likes", 0),
"downloads": item.get("downloads", 0),
"is_nc": item.get("is_nc", False),
"new": item.get("new", False),
}
for item in data
]
with open("defaults_data.json", "r") as file:
lora_defaults = json.load(file)
device = "cuda"
state_dicts = {}
for item in sdxl_loras_raw:
saved_name = hf_hub_download(item["repo"], item["weights"])
if not saved_name.endswith('.safetensors'):
state_dict = torch.load(saved_name)
else:
state_dict = load_file(saved_name)
state_dicts[item["repo"]] = {
"saved_name": saved_name,
"state_dict": state_dict
}
sdxl_loras_raw = [item for item in sdxl_loras_raw if item.get("new") != True]
# download models
hf_hub_download(
repo_id="InstantX/InstantID",
filename="ControlNetModel/config.json",
local_dir="/data/checkpoints",
)
hf_hub_download(
repo_id="InstantX/InstantID",
filename="ControlNetModel/diffusion_pytorch_model.safetensors",
local_dir="/data/checkpoints",
)
hf_hub_download(
repo_id="InstantX/InstantID", filename="ip-adapter.bin", local_dir="/data/checkpoints"
)
hf_hub_download(
repo_id="latent-consistency/lcm-lora-sdxl",
filename="pytorch_lora_weights.safetensors",
local_dir="/data/checkpoints",
)
# download antelopev2
if not os.path.exists("/data/antelopev2.zip"):
gdown.download(url="https://drive.google.com/file/d/18wEUfMNohBJ4K3Ly5wpTejPfDzp-8fI8/view?usp=sharing", output="/data/", quiet=False, fuzzy=True)
os.system("unzip /data/antelopev2.zip -d /data/models/")
app = FaceAnalysis(name='antelopev2', root='/data', providers=['CPUExecutionProvider'])
app.prepare(ctx_id=0, det_size=(640, 640))
# prepare models under ./checkpoints
face_adapter = f'/data/checkpoints/ip-adapter.bin'
controlnet_path = f'/data/checkpoints/ControlNetModel'
# load IdentityNet
st = time.time()
identitynet = ControlNetModel.from_pretrained(controlnet_path, torch_dtype=torch.float16)
zoedepthnet = ControlNetModel.from_pretrained("diffusers/controlnet-zoe-depth-sdxl-1.0",torch_dtype=torch.float16)
et = time.time()
elapsed_time = et - st
print('Loading ControlNet took: ', elapsed_time, 'seconds')
st = time.time()
vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
et = time.time()
elapsed_time = et - st
print('Loading VAE took: ', elapsed_time, 'seconds')
st = time.time()
pipe = StableDiffusionXLInstantIDImg2ImgPipeline.from_pretrained("rubbrband/albedobaseXL_v21",
vae=vae,
controlnet=[identitynet, zoedepthnet],
torch_dtype=torch.float16)
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)
pipe.load_ip_adapter_instantid(face_adapter)
pipe.set_ip_adapter_scale(0.8)
et = time.time()
elapsed_time = et - st
print('Loading pipeline took: ', elapsed_time, 'seconds')
st = time.time()
compel = Compel(tokenizer=[pipe.tokenizer, pipe.tokenizer_2] , text_encoder=[pipe.text_encoder, pipe.text_encoder_2], returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED, requires_pooled=[False, True])
et = time.time()
elapsed_time = et - st
print('Loading Compel took: ', elapsed_time, 'seconds')
st = time.time()
zoe = ZoeDetector.from_pretrained("lllyasviel/Annotators")
et = time.time()
elapsed_time = et - st
print('Loading Zoe took: ', elapsed_time, 'seconds')
zoe.to(device)
pipe.to(device)
last_lora = ""
last_fused = False
js = '''
var button = document.getElementById('button');
// Add a click event listener to the button
button.addEventListener('click', function() {
element.classList.add('selected');
});
'''
def update_selection(selected_state: gr.SelectData, sdxl_loras, face_strength, image_strength, weight, depth_control_scale, negative, is_new=False):
lora_repo = sdxl_loras[selected_state.index]["repo"]
new_placeholder = "Type a prompt to use your selected LoRA"
weight_name = sdxl_loras[selected_state.index]["weights"]
updated_text = f"### Selected: [{lora_repo}](https://huggingface.co/{lora_repo}) ✨ {'(non-commercial LoRA, `cc-by-nc`)' if sdxl_loras[selected_state.index]['is_nc'] else '' }"
for lora_list in lora_defaults:
if lora_list["model"] == sdxl_loras[selected_state.index]["repo"]:
face_strength = lora_list.get("face_strength", 0.85)
image_strength = lora_list.get("image_strength", 0.15)
weight = lora_list.get("weight", 0.9)
depth_control_scale = lora_list.get("depth_control_scale", 0.8)
negative = lora_list.get("negative", "")
if(is_new):
if(selected_state.index == 0):
selected_state.index = -9999
else:
selected_state.index *= -1
return (
updated_text,
gr.update(placeholder=new_placeholder),
face_strength,
image_strength,
weight,
depth_control_scale,
negative,
selected_state
)
def center_crop_image_as_square(img):
square_size = min(img.size)
left = (img.width - square_size) / 2
top = (img.height - square_size) / 2
right = (img.width + square_size) / 2
bottom = (img.height + square_size) / 2
img_cropped = img.crop((left, top, right, bottom))
return img_cropped
def check_selected(selected_state):
if not selected_state:
raise gr.Error("You must select a style")
def merge_incompatible_lora(full_path_lora, lora_scale):
for weights_file in [full_path_lora]:
if ";" in weights_file:
weights_file, multiplier = weights_file.split(";")
multiplier = float(multiplier)
else:
multiplier = lora_scale
lora_model, weights_sd = lora.create_network_from_weights(
multiplier,
full_path_lora,
pipe.vae,
pipe.text_encoder,
pipe.unet,
for_inference=True,
)
lora_model.merge_to(
pipe.text_encoder, pipe.unet, weights_sd, torch.float16, "cuda"
)
del weights_sd
del lora_model
@spaces.GPU
def generate_image(prompt, negative, face_emb, face_image, face_kps, image_strength, guidance_scale, face_strength, depth_control_scale, repo_name, loaded_state_dict, lora_scale, sdxl_loras, selected_state_index, st):
et = time.time()
elapsed_time = et - st
print('Getting into the decorated function took: ', elapsed_time, 'seconds')
global last_fused, last_lora
print("Last LoRA: ", last_lora)
print("Current LoRA: ", repo_name)
print("Last fused: ", last_fused)
#prepare face zoe
st = time.time()
with torch.no_grad():
image_zoe = zoe(face_image)
width, height = face_kps.size
images = [face_kps, image_zoe.resize((height, width))]
et = time.time()
elapsed_time = et - st
print('Zoe Depth calculations took: ', elapsed_time, 'seconds')
if last_lora != repo_name:
if(last_fused):
st = time.time()
pipe.unfuse_lora()
pipe.unload_lora_weights()
et = time.time()
elapsed_time = et - st
print('Unfuse and unload LoRA took: ', elapsed_time, 'seconds')
st = time.time()
pipe.load_lora_weights(loaded_state_dict)
pipe.fuse_lora(lora_scale)
et = time.time()
elapsed_time = et - st
print('Fuse and load LoRA took: ', elapsed_time, 'seconds')
last_fused = True
is_pivotal = sdxl_loras[selected_state_index]["is_pivotal"]
if(is_pivotal):
#Add the textual inversion embeddings from pivotal tuning models
text_embedding_name = sdxl_loras[selected_state_index]["text_embedding_weights"]
embedding_path = hf_hub_download(repo_id=repo_name, filename=text_embedding_name, repo_type="model")
state_dict_embedding = load_file(embedding_path)
try:
pipe.unload_textual_inversion()
pipe.load_textual_inversion(state_dict_embedding["clip_l"], token=["<s0>", "<s1>"], text_encoder=pipe.text_encoder, tokenizer=pipe.tokenizer)
pipe.load_textual_inversion(state_dict_embedding["clip_g"], token=["<s0>", "<s1>"], text_encoder=pipe.text_encoder_2, tokenizer=pipe.tokenizer_2)
except:
pipe.unload_textual_inversion()
pipe.load_textual_inversion(state_dict_embedding["text_encoders_0"], token=["<s0>", "<s1>"], text_encoder=pipe.text_encoder, tokenizer=pipe.tokenizer)
pipe.load_textual_inversion(state_dict_embedding["text_encoders_1"], token=["<s0>", "<s1>"], text_encoder=pipe.text_encoder_2, tokenizer=pipe.tokenizer_2)
print("Processing prompt...")
st = time.time()
conditioning, pooled = compel(prompt)
if(negative):
negative_conditioning, negative_pooled = compel(negative)
else:
negative_conditioning, negative_pooled = None, None
et = time.time()
elapsed_time = et - st
print('Prompt processing took: ', elapsed_time, 'seconds')
print("Processing image...")
st = time.time()
image = pipe(
prompt_embeds=conditioning,
pooled_prompt_embeds=pooled,
negative_prompt_embeds=negative_conditioning,
negative_pooled_prompt_embeds=negative_pooled,
width=1024,
height=1024,
image_embeds=face_emb,
image=face_image,
strength=1-image_strength,
control_image=images,
num_inference_steps=20,
guidance_scale = guidance_scale,
controlnet_conditioning_scale=[face_strength, depth_control_scale],
).images[0]
et = time.time()
elapsed_time = et - st
print('Image processing took: ', elapsed_time, 'seconds')
last_lora = repo_name
return image
def run_lora(face_image, prompt, negative, lora_scale, selected_state, face_strength, image_strength, guidance_scale, depth_control_scale, sdxl_loras, progress=gr.Progress(track_tqdm=True)):
selected_state_index = selected_state.index
st = time.time()
face_image = center_crop_image_as_square(face_image)
try:
face_info = app.get(cv2.cvtColor(np.array(face_image), cv2.COLOR_RGB2BGR))
face_info = sorted(face_info, key=lambda x:(x['bbox'][2]-x['bbox'][0])*x['bbox'][3]-x['bbox'][1])[-1] # only use the maximum face
face_emb = face_info['embedding']
face_kps = draw_kps(face_image, face_info['kps'])
except:
raise gr.Error("No face found in your image. Only face images work here. Try again")
et = time.time()
elapsed_time = et - st
print('Cropping and calculating face embeds took: ', elapsed_time, 'seconds')
st = time.time()
for lora_list in lora_defaults:
if lora_list["model"] == sdxl_loras[selected_state_index]["repo"]:
prompt_full = lora_list.get("prompt", None)
if(prompt_full):
prompt = prompt_full.replace("<subject>", prompt)
print("Prompt:", prompt)
if(prompt == ""):
prompt = "a person"
print("Selected State: ", selected_state_index)
print(sdxl_loras[selected_state_index]["repo"])
if negative == "":
negative = None
if not selected_state:
raise gr.Error("You must select a LoRA")
repo_name = sdxl_loras[selected_state_index]["repo"]
weight_name = sdxl_loras[selected_state_index]["weights"]
full_path_lora = state_dicts[repo_name]["saved_name"]
#loaded_state_dict = copy.deepcopy(state_dicts[repo_name]["state_dict"])
cross_attention_kwargs = None
et = time.time()
elapsed_time = et - st
print('Small content processing took: ', elapsed_time, 'seconds')
st = time.time()
image = generate_image(prompt, negative, face_emb, face_image, face_kps, image_strength, guidance_scale, face_strength, depth_control_scale, repo_name, full_path_lora, lora_scale, sdxl_loras, selected_state_index, st)
return image, gr.update(visible=True)
def shuffle_gallery(sdxl_loras):
random.shuffle(sdxl_loras)
return [(item["image"], item["title"]) for item in sdxl_loras], sdxl_loras
def classify_gallery(sdxl_loras):
sorted_gallery = sorted(sdxl_loras, key=lambda x: x.get("likes", 0), reverse=True)
return [(item["image"], item["title"]) for item in sorted_gallery], sorted_gallery
def swap_gallery(order, sdxl_loras):
if(order == "random"):
return shuffle_gallery(sdxl_loras)
else:
return classify_gallery(sdxl_loras)
def deselect():
return gr.Gallery(selected_index=None)
with gr.Blocks(css="custom.css") as demo:
gr_sdxl_loras = gr.State(value=sdxl_loras_raw)
title = gr.HTML(
"""<h1><img src="https://i.imgur.com/DVoGw04.png">
<span>Face to All<br><small style="
font-size: 13px;
display: block;
font-weight: normal;
opacity: 0.75;
">🧨 diffusers InstantID + ControlNet<br> inspired by fofr's <a href="https://github.com/fofr/cog-face-to-many" target="_blank">face-to-many</a></small></span></h1>""",
elem_id="title",
)
selected_state = gr.State()
with gr.Row(elem_id="main_app"):
with gr.Column(scale=4):
with gr.Group(elem_id="gallery_box"):
photo = gr.Image(label="Upload a picture of yourself", interactive=True, type="pil", height=300)
selected_loras = gr.Gallery(label="Selected LoRAs", height=80, show_share_button=False, visible=False, elem_id="gallery_selected", )
#order_gallery = gr.Radio(choices=["random", "likes"], value="random", label="Order by", elem_id="order_radio")
#new_gallery = gr.Gallery(
# label="New LoRAs",
# elem_id="gallery_new",
# columns=3,
# value=[(item["image"], item["title"]) for item in sdxl_loras_raw_new], allow_preview=False, show_share_button=False)
gallery = gr.Gallery(
#value=[(item["image"], item["title"]) for item in sdxl_loras],
label="Style gallery",
allow_preview=False,
columns=4,
elem_id="gallery",
show_share_button=False,
height=550
)
custom_model = gr.Textbox(label="Enter a custom Hugging Face or CivitAI SDXL LoRA", interactive=False, placeholder="Coming soon...")
with gr.Column(scale=5):
with gr.Row():
prompt = gr.Textbox(label="Prompt", show_label=False, lines=1, max_lines=1, info="Describe your subject (optional)", value="a person", elem_id="prompt")
button = gr.Button("Run", elem_id="run_button")
result = gr.Image(
interactive=False, label="Generated Image", elem_id="result-image"
)
with gr.Group(elem_id="share-btn-container", visible=False) as share_group:
community_icon = gr.HTML(community_icon_html)
loading_icon = gr.HTML(loading_icon_html)
share_button = gr.Button("Share to community", elem_id="share-btn")
with gr.Accordion("Advanced options", open=False):
negative = gr.Textbox(label="Negative Prompt")
weight = gr.Slider(0, 10, value=0.9, step=0.1, label="LoRA weight")
face_strength = gr.Slider(0, 1, value=0.85, step=0.01, label="Face strength", info="Higher values increase the face likeness but reduce the creative liberty of the models")
image_strength = gr.Slider(0, 1, value=0.15, step=0.01, label="Image strength", info="Higher values increase the similarity with the structure/colors of the original photo")
guidance_scale = gr.Slider(0, 50, value=7, step=0.1, label="Guidance Scale")
depth_control_scale = gr.Slider(0, 1, value=0.8, step=0.01, label="Zoe Depth ControlNet strenght")
prompt_title = gr.Markdown(
value="### Click on a LoRA in the gallery to select it",
visible=True,
elem_id="selected_lora",
)
#order_gallery.change(
# fn=swap_gallery,
# inputs=[order_gallery, gr_sdxl_loras],
# outputs=[gallery, gr_sdxl_loras],
# queue=False
#)
gallery.select(
fn=update_selection,
inputs=[gr_sdxl_loras, face_strength, image_strength, weight, depth_control_scale, negative],
outputs=[prompt_title, prompt, face_strength, image_strength, weight, depth_control_scale, negative, selected_state],
queue=False,
show_progress=False
)
#new_gallery.select(
# fn=update_selection,
# inputs=[gr_sdxl_loras_new, gr.State(True)],
# outputs=[prompt_title, prompt, prompt, selected_state, gallery],
# queue=False,
# show_progress=False
#)
prompt.submit(
fn=check_selected,
inputs=[selected_state],
queue=False,
show_progress=False
).success(
fn=run_lora,
inputs=[photo, prompt, negative, weight, selected_state, face_strength, image_strength, guidance_scale, depth_control_scale, gr_sdxl_loras],
outputs=[result, share_group],
)
button.click(
fn=check_selected,
inputs=[selected_state],
queue=False,
show_progress=False
).success(
fn=run_lora,
inputs=[photo, prompt, negative, weight, selected_state, face_strength, image_strength, guidance_scale, depth_control_scale, gr_sdxl_loras],
outputs=[result, share_group],
)
share_button.click(None, [], [], js=share_js)
demo.load(fn=classify_gallery, inputs=[gr_sdxl_loras], outputs=[gallery, gr_sdxl_loras], queue=False, js=js)
demo.queue(max_size=20)
demo.launch(share=True) |