File size: 12,569 Bytes
a6c206d |
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 |
import gradio as gr
import numpy as np
import random
import spaces
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler, AutoencoderTiny, AutoencoderKL
from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5TokenizerFast
from model import Flux
def calculate_shift(
image_seq_len,
base_seq_len: int = 256,
max_seq_len: int = 4096,
base_shift: float = 0.5,
max_shift: float = 1.16,
):
m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
b = base_shift - m * base_seq_len
mu = image_seq_len * m + b
return mu
def retrieve_timesteps(
scheduler,
num_inference_steps: Optional[int] = None,
device: Optional[Union[str, torch.device]] = None,
timesteps: Optional[List[int]] = None,
sigmas: Optional[List[float]] = None,
**kwargs,
):
if timesteps is not None and sigmas is not None:
raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
if timesteps is not None:
scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
timesteps = scheduler.timesteps
num_inference_steps = len(timesteps)
elif sigmas is not None:
scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
timesteps = scheduler.timesteps
num_inference_steps = len(timesteps)
else:
scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
timesteps = scheduler.timesteps
return timesteps, num_inference_steps
@torch.inference_mode()
def flux_pipe_call_that_returns_an_iterable_of_images(
self,
prompt: Union[str, List[str]] = None,
prompt_2: Optional[Union[str, List[str]]] = None,
height: Optional[int] = None,
width: Optional[int] = None,
num_inference_steps: int = 28,
timesteps: List[int] = None,
guidance_scale: float = 3.5,
num_images_per_prompt: Optional[int] = 1,
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
latents: Optional[torch.FloatTensor] = None,
prompt_embeds: Optional[torch.FloatTensor] = None,
pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
output_type: Optional[str] = "pil",
return_dict: bool = True,
joint_attention_kwargs: Optional[Dict[str, Any]] = None,
max_sequence_length: int = 512,
good_vae: Optional[Any] = None,
):
height = height or self.default_sample_size * self.vae_scale_factor
width = width or self.default_sample_size * self.vae_scale_factor
# 1. Check inputs
self.check_inputs(
prompt,
prompt_2,
height,
width,
prompt_embeds=prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
max_sequence_length=max_sequence_length,
)
self._guidance_scale = guidance_scale
self._joint_attention_kwargs = joint_attention_kwargs
self._interrupt = False
# 2. Define call parameters
batch_size = 1 if isinstance(prompt, str) else len(prompt)
device = self._execution_device
# 3. Encode prompt
lora_scale = joint_attention_kwargs.get("scale", None) if joint_attention_kwargs is not None else None
prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(
prompt=prompt,
prompt_2=prompt_2,
prompt_embeds=prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
device=device,
num_images_per_prompt=num_images_per_prompt,
max_sequence_length=max_sequence_length,
lora_scale=lora_scale,
)
# 4. Prepare latent variables
num_channels_latents = self.transformer.config.in_channels // 4
latents, latent_image_ids = self.prepare_latents(
batch_size * num_images_per_prompt,
num_channels_latents,
height,
width,
prompt_embeds.dtype,
device,
generator,
latents,
)
# 5. Prepare timesteps
sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
image_seq_len = latents.shape[1]
mu = calculate_shift(
image_seq_len,
self.scheduler.config.base_image_seq_len,
self.scheduler.config.max_image_seq_len,
self.scheduler.config.base_shift,
self.scheduler.config.max_shift,
)
timesteps, num_inference_steps = retrieve_timesteps(
self.scheduler,
num_inference_steps,
device,
timesteps,
sigmas,
mu=mu,
)
self._num_timesteps = len(timesteps)
# Handle guidance
guidance = torch.full([1], guidance_scale, device=device, dtype=torch.float32).expand(latents.shape[0]) if self.transformer.config.guidance_embeds else None
# 6. Denoising loop
for i, t in enumerate(timesteps):
if self.interrupt:
continue
timestep = t.expand(latents.shape[0]).to(latents.dtype)
noise_pred = self.transformer(
hidden_states=latents,
timestep=timestep / 1000,
guidance=guidance,
pooled_projections=pooled_prompt_embeds,
encoder_hidden_states=prompt_embeds,
txt_ids=text_ids,
img_ids=latent_image_ids,
joint_attention_kwargs=self.joint_attention_kwargs,
return_dict=False,
)[0]
# Yield intermediate result
latents_for_image = self._unpack_latents(latents, height, width, self.vae_scale_factor)
latents_for_image = (latents_for_image / self.vae.config.scaling_factor) + self.vae.config.shift_factor
image = self.vae.decode(latents_for_image, return_dict=False)[0]
yield self.image_processor.postprocess(image, output_type=output_type)[0]
latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
torch.cuda.empty_cache()
# Final image using good_vae
latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
latents = (latents / good_vae.config.scaling_factor) + good_vae.config.shift_factor
image = good_vae.decode(latents, return_dict=False)[0]
self.maybe_free_model_hooks()
torch.cuda.empty_cache()
yield self.image_processor.postprocess(image, output_type=output_type)[0]
@dataclass
class ModelSpec:
params: FluxParams
ae_params: AutoEncoderParams
ckpt_path: str
ae_path: str
repo_id: str
repo_flow: str
repo_ae: str
repo_id_ae: str
config = ModelSpec(
repo_id="TencentARC/flux-mini",
repo_flow="flux-mini.safetensors",
repo_id_ae="black-forest-labs/FLUX.1-dev",
repo_ae="ae.safetensors",
ckpt_path=os.getenv("FLUX_MINI", None),
params=FluxParams(
in_channels=64,
vec_in_dim=768,
context_in_dim=4096,
hidden_size=3072,
mlp_ratio=4.0,
num_heads=24,
depth=5,
depth_single_blocks=10,
axes_dim=[16, 56, 56],
theta=10_000,
qkv_bias=True,
guidance_embed=True,
)
def load_flow_model2(device: str = "cuda", hf_download: bool = True):
if (
and config.repo_id is not None
and config.repo_flow is not None
and hf_download
):
ckpt_path = hf_hub_download(configs[name].repo_id, configs[name].repo_flow.replace("sft", "safetensors"))
model = Flux(params)
if ckpt_path is not None:
sd = load_sft(ckpt_path, device=str(device))
missing, unexpected = model.load_state_dict(sd, strict=True)
return model
dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"
scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="scheduler").to(device)
vae = AutoencoderKL.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="vae", torch_dtype=dtype).to(device)
text_encoder = CLIPTextModel.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="text_encoder").to(device)
tokenizer = CLIPTokenizer.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="tokenizer").to(device)
text_encoder_2 = T5EncoderModel.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="text_encoder_2").to(device)
tokenizer_2 = T5TokenizerFast.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="tokenizer_2").to(device)
transformer = load_flow_model2(device)
pipe = FluxPipeline(
scheduler,
vae,
text_encoder,
tokenizer,
text_encoder_2,
tokenizer_2
transformer
)
torch.cuda.empty_cache()
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048
pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)
@spaces.GPU(duration=75)
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=3.5, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)):
if randomize_seed:
seed = random.randint(0, MAX_SEED)
generator = torch.Generator().manual_seed(seed)
for img in pipe.flux_pipe_call_that_returns_an_iterable_of_images(
prompt=prompt,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
width=width,
height=height,
generator=generator,
output_type="pil",
good_vae=good_vae,
):
yield img, seed
examples = [
"thousands of luminous oysters on a shore reflecting and refracting the sunset",
"profile of sad Socrates, full body, high detail, dramatic scene, Epic dynamic action, wide angle, cinematic, hyper realistic, concept art, warm muted tones as painted by Bernie Wrightson, Frank Frazetta,",
"ghosts, astronauts, robots, cats, superhero costumes, line drawings, naive, simple, exploring a strange planet, coloured pencil crayons, , black canvas background, drawn by 5 year old child",
]
css="""
#col-container {
margin: 0 auto;
max-width: 520px;
}
"""
with gr.Blocks(css=css) as demo:
with gr.Column(elem_id="col-container"):
gr.Markdown(f"""# FLUX-Mini
A 3.2B param rectified flow transformer distilled from [FLUX.1 [dev]](https://blackforestlabs.ai/)
[[non-commercial license](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md)]
""")
with gr.Row():
prompt = gr.Text(
label="Prompt",
show_label=False,
max_lines=1,
placeholder="Enter your prompt",
container=False,
)
run_button = gr.Button("Run", scale=0)
result = gr.Image(label="Result", show_label=False)
with gr.Accordion("Advanced Settings", open=False):
seed = gr.Slider(
label="Seed",
minimum=0,
maximum=MAX_SEED,
step=1,
value=0,
)
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
with gr.Row():
width = gr.Slider(
label="Width",
minimum=256,
maximum=MAX_IMAGE_SIZE,
step=32,
value=1024,
)
height = gr.Slider(
label="Height",
minimum=256,
maximum=MAX_IMAGE_SIZE,
step=32,
value=1024,
)
with gr.Row():
guidance_scale = gr.Slider(
label="Guidance Scale",
minimum=1,
maximum=15,
step=0.1,
value=3.5,
)
num_inference_steps = gr.Slider(
label="Number of inference steps",
minimum=1,
maximum=50,
step=1,
value=28,
)
gr.Examples(
examples = examples,
fn = infer,
inputs = [prompt],
outputs = [result, seed],
cache_examples="lazy"
)
gr.on(
triggers=[run_button.click, prompt.submit],
fn = infer,
inputs = [prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
outputs = [result, seed]
)
demo.launch() |