File size: 9,009 Bytes
e9af91e |
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 |
import spaces
import torch
import random
import numpy as np
from inspect import signature
from diffusers import (
FluxPipeline,
StableDiffusion3Pipeline,
PixArtSigmaPipeline,
SanaPipeline,
AuraFlowPipeline,
Kandinsky3Pipeline,
HunyuanDiTPipeline,
LuminaText2ImgPipeline,AutoPipelineForText2Image
)
import gradio as gr
from diffusers.pipelines.pipeline_utils import DiffusionPipeline
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024
class ProgressPipeline(DiffusionPipeline):
def __init__(self, original_pipeline):
super().__init__()
self.original_pipeline = original_pipeline
# Register all components from the original pipeline
for attr_name, attr_value in vars(original_pipeline).items():
setattr(self, attr_name, attr_value)
@torch.no_grad()
def __call__(
self,
prompt,
num_inference_steps=30,
generator=None,
guidance_scale=7.5,
callback=None,
callback_steps=1,
**kwargs
):
# Initialize the progress tracking
self._num_inference_steps = num_inference_steps
self._step = 0
def progress_callback(step_index, timestep, callback_kwargs):
if callback and step_index % callback_steps == 0:
# Pass self (the pipeline) to the callback
callback(self, step_index, timestep, callback_kwargs)
return callback_kwargs
# Monkey patch the original pipeline's progress tracking
original_step = self.original_pipeline.scheduler.step
def wrapped_step(*args, **kwargs):
self._step += 1
progress_callback(self._step, None, {})
return original_step(*args, **kwargs)
self.original_pipeline.scheduler.step = wrapped_step
try:
# Call the original pipeline
result = self.original_pipeline(
prompt=prompt,
num_inference_steps=num_inference_steps,
generator=generator,
guidance_scale=guidance_scale,
**kwargs
)
return result
finally:
# Restore the original step function
self.original_pipeline.scheduler.step = original_step
cache_dir = '/workspace/hf_cache'
MODEL_CONFIGS = {
"FLUX": {
"repo_id": "black-forest-labs/FLUX.1-dev",
"pipeline_class": FluxPipeline,
},
"Stable Diffusion 3.5": {
"repo_id": "stabilityai/stable-diffusion-3.5-large",
"pipeline_class": StableDiffusion3Pipeline,
},
"PixArt": {
"repo_id": "PixArt-alpha/PixArt-Sigma-XL-2-1024-MS",
"pipeline_class": PixArtSigmaPipeline,
},
"SANA": {
"repo_id": "Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers",
"pipeline_class": SanaPipeline,
},
"AuraFlow": {
"repo_id": "fal/AuraFlow",
"pipeline_class": AuraFlowPipeline,
},
"Kandinsky": {
"repo_id": "kandinsky-community/kandinsky-3",
"pipeline_class": Kandinsky3Pipeline,
},
"Hunyuan": {
"repo_id": "Tencent-Hunyuan/HunyuanDiT-Diffusers",
"pipeline_class": HunyuanDiTPipeline,
},
"Lumina": {
"repo_id": "Alpha-VLLM/Lumina-Next-SFT-diffusers",
"pipeline_class": LuminaText2ImgPipeline,
}
}
def generate_image_with_progress(model_name,pipe, prompt, num_steps, guidance_scale=3.5, seed=None,negative_prompt=None, randomize_seed=None, width=1024, height=1024, num_inference_steps=40, progress=gr.Progress(track_tqdm=True)):
generator = None
if randomize_seed:
seed = random.randint(0, MAX_SEED)
if seed is not None:
generator = torch.Generator("cuda").manual_seed(seed)
else:
generator = torch.Generator("cuda")
def callback(pipe, step_index, timestep, callback_kwargs):
print(f" callback => {step_index}, {timestep}")
if step_index is None:
step_index = 0
cur_prg = step_index / num_steps
progress(cur_prg, desc=f"Step {step_index}/{num_steps}")
return callback_kwargs
print(f"START GENR ")
# Get the signature of the pipe
pipe_signature = signature(pipe)
# Check for the presence of "guidance_scale" and "callback_on_step_end" in the signature
has_guidance_scale = "guidance_scale" in pipe_signature.parameters
has_callback_on_step_end = "callback_on_step_end" in pipe_signature.parameters
# Define common arguments
common_args = {
"prompt": prompt,
"num_inference_steps": num_steps,
"negative_prompt": negative_prompt,
"width": width,
"height": height,
"generator": generator,
}
if has_guidance_scale:
common_args["guidance_scale"] = guidance_scale
if has_callback_on_step_end:
print("has callback_on_step_end and", "has guidance_scale" if has_guidance_scale else "NO guidance_scale")
common_args["callback_on_step_end"] = callback
else:
print("NO callback_on_step_end and", "has guidance_scale" if has_guidance_scale else "NO guidance_scale")
common_args["callback"] = callback
common_args["callback_steps"] = 1
# Generate image
image = pipe(**common_args).images[0]
return seed, image
@spaces.GPU(duration=170)
def create_pipeline_logic(prompt_text, model_name, negative_prompt="", seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=4.5, num_inference_steps=40,):
print(f"starting {model_name}")
progress = gr.Progress(track_tqdm=True)
config = MODEL_CONFIGS[model_name]
pipe_class = config["pipeline_class"]
pipe = None
b_pipe = AutoPipelineForText2Image.from_pretrained(
config["repo_id"],
#variant="fp16",
#cache_dir=config["cache_dir"],
torch_dtype=torch.bfloat16
).to("cuda")
pipe_signature = signature(b_pipe)
# Check for the presence of "callback_on_step_end" in the signature
has_callback_on_step_end = "callback_on_step_end" in pipe_signature.parameters
if not has_callback_on_step_end:
pipe = ProgressPipeline(b_pipe)
print("ProgressPipeline specal")
else:
pipe = b_pipe
gen_seed,image = generate_image_with_progress(
model_name,pipe, prompt_text, num_steps=num_inference_steps, guidance_scale=guidance_scale, seed=seed,negative_prompt = negative_prompt, randomize_seed = randomize_seed, width = width, height = height, progress=progress
)
return f"Seed: {gen_seed}", image
def main():
with gr.Blocks() as app:
gr.Markdown("# Dynamic Multiple Model Image Generation")
prompt_text = gr.Textbox(label="Enter prompt")
with gr.Accordion("Advanced Settings", open=False):
negative_prompt = gr.Text(
label="Negative prompt",
max_lines=1,
placeholder="Enter a negative prompt",
)
seed = gr.Slider(
label="Seed",
minimum=0,
maximum=MAX_SEED,
step=100,
value=0,
)
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
with gr.Row():
width = gr.Slider(
label="Width",
minimum=512,
maximum=MAX_IMAGE_SIZE,
step=32,
value=1024,
)
height = gr.Slider(
label="Height",
minimum=512,
maximum=MAX_IMAGE_SIZE,
step=32,
value=1024,
)
with gr.Row():
guidance_scale = gr.Slider(
label="Guidance scale",
minimum=0.0,
maximum=7.5,
step=0.1,
value=4.5,
)
num_inference_steps = gr.Slider(
label="Number of inference steps",
minimum=1,
maximum=50,
step=1,
value=40,
)
for model_name, config in MODEL_CONFIGS.items():
with gr.Tab(model_name):
button = gr.Button(f"Run {model_name}")
output = gr.Textbox(label="Status")
img = gr.Image(label=model_name, height=300)
button.click(fn=create_pipeline_logic, inputs=[prompt_text, gr.Text(value= model_name,visible=False), negative_prompt,
seed,
randomize_seed,
width,
height,
guidance_scale,
num_inference_steps], outputs=[output, img])
app.launch()
if __name__ == "__main__":
main()
|