Spaces:
Running
on
Zero
Running
on
Zero
ui revamp
Browse files
app.py
CHANGED
|
@@ -139,14 +139,43 @@ def upsample_prompt_logic(prompt, image_list):
|
|
| 139 |
print(f"Upsampling failed: {e}")
|
| 140 |
return prompt
|
| 141 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 142 |
# Updated duration function to match generate_image arguments (including progress)
|
| 143 |
-
def get_duration(prompt_embeds, image_list, width, height, num_inference_steps, guidance_scale, seed,
|
| 144 |
num_images = 0 if image_list is None else len(image_list)
|
| 145 |
step_duration = 1 + 0.8 * num_images
|
| 146 |
return max(65, num_inference_steps * step_duration + 10)
|
| 147 |
|
| 148 |
@spaces.GPU(duration=get_duration)
|
| 149 |
-
def generate_image(prompt_embeds, image_list, width, height, num_inference_steps, guidance_scale, seed,
|
| 150 |
# Move embeddings to GPU only when inside the GPU decorated function
|
| 151 |
prompt_embeds = prompt_embeds.to(device)
|
| 152 |
|
|
@@ -158,12 +187,10 @@ def generate_image(prompt_embeds, image_list, width, height, num_inference_steps
|
|
| 158 |
"num_inference_steps": num_inference_steps,
|
| 159 |
"guidance_scale": guidance_scale,
|
| 160 |
"generator": generator,
|
|
|
|
|
|
|
| 161 |
}
|
| 162 |
|
| 163 |
-
if image_list is None or force_dimensions:
|
| 164 |
-
pipe_kwargs["width"] = width
|
| 165 |
-
pipe_kwargs["height"] = height
|
| 166 |
-
|
| 167 |
# Progress bar for the actual generation steps
|
| 168 |
if progress:
|
| 169 |
progress(0, desc="Starting generation...")
|
|
@@ -171,7 +198,7 @@ def generate_image(prompt_embeds, image_list, width, height, num_inference_steps
|
|
| 171 |
image = pipe(**pipe_kwargs).images[0]
|
| 172 |
return image
|
| 173 |
|
| 174 |
-
def infer(prompt, input_images=None, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=50, guidance_scale=2.5,
|
| 175 |
|
| 176 |
if randomize_seed:
|
| 177 |
seed = random.randint(0, MAX_SEED)
|
|
@@ -206,7 +233,6 @@ def infer(prompt, input_images=None, seed=42, randomize_seed=False, width=1024,
|
|
| 206 |
num_inference_steps,
|
| 207 |
guidance_scale,
|
| 208 |
seed,
|
| 209 |
-
force_dimensions,
|
| 210 |
progress
|
| 211 |
)
|
| 212 |
|
|
@@ -227,7 +253,7 @@ examples_images = [
|
|
| 227 |
css="""
|
| 228 |
#col-container {
|
| 229 |
margin: 0 auto;
|
| 230 |
-
max-width:
|
| 231 |
}
|
| 232 |
.gallery-container img{
|
| 233 |
object-fit: contain;
|
|
@@ -240,89 +266,85 @@ with gr.Blocks() as demo:
|
|
| 240 |
gr.Markdown(f"""# FLUX.2 [dev]
|
| 241 |
FLUX.2 [dev] is a 32B model rectified flow capable of generating, editing and combining images based on text instructions model [[model](https://huggingface.co/black-forest-labs/FLUX.2-dev)], [[blog](https://bfl.ai/blog/flux-2)]
|
| 242 |
""")
|
| 243 |
-
|
| 244 |
-
with gr.Accordion("Input image(s) (optional)", open=True):
|
| 245 |
-
input_images = gr.Gallery(
|
| 246 |
-
label="Input Image(s)",
|
| 247 |
-
type="pil",
|
| 248 |
-
columns=3,
|
| 249 |
-
rows=1,
|
| 250 |
-
)
|
| 251 |
-
|
| 252 |
with gr.Row():
|
| 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 |
-
width = gr.Slider(
|
| 288 |
-
label="Width",
|
| 289 |
-
minimum=256,
|
| 290 |
-
maximum=MAX_IMAGE_SIZE,
|
| 291 |
-
step=32,
|
| 292 |
-
value=1024,
|
| 293 |
-
)
|
| 294 |
|
| 295 |
-
|
| 296 |
-
|
| 297 |
-
minimum=256,
|
| 298 |
-
maximum=MAX_IMAGE_SIZE,
|
| 299 |
-
step=32,
|
| 300 |
-
value=1024,
|
| 301 |
-
)
|
| 302 |
|
| 303 |
-
force_dimensions = gr.Checkbox(
|
| 304 |
-
label="Force width/height when image input",
|
| 305 |
-
value=False,
|
| 306 |
-
info="When unchecked, width/height settings are ignored if input images are provided"
|
| 307 |
-
)
|
| 308 |
-
|
| 309 |
-
with gr.Row():
|
| 310 |
-
|
| 311 |
-
num_inference_steps = gr.Slider(
|
| 312 |
-
label="Number of inference steps",
|
| 313 |
-
minimum=1,
|
| 314 |
-
maximum=100,
|
| 315 |
-
step=1,
|
| 316 |
-
value=30,
|
| 317 |
-
)
|
| 318 |
-
|
| 319 |
-
guidance_scale = gr.Slider(
|
| 320 |
-
label="Guidance scale",
|
| 321 |
-
minimum=0.0,
|
| 322 |
-
maximum=10.0,
|
| 323 |
-
step=0.1,
|
| 324 |
-
value=4,
|
| 325 |
-
)
|
| 326 |
|
| 327 |
gr.Examples(
|
| 328 |
examples=examples,
|
|
@@ -342,10 +364,17 @@ FLUX.2 [dev] is a 32B model rectified flow capable of generating, editing and co
|
|
| 342 |
cache_mode="lazy"
|
| 343 |
)
|
| 344 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 345 |
gr.on(
|
| 346 |
triggers=[run_button.click, prompt.submit],
|
| 347 |
fn=infer,
|
| 348 |
-
inputs=[prompt, input_images, seed, randomize_seed, width, height, num_inference_steps, guidance_scale,
|
| 349 |
outputs=[result, seed]
|
| 350 |
)
|
| 351 |
|
|
|
|
| 139 |
print(f"Upsampling failed: {e}")
|
| 140 |
return prompt
|
| 141 |
|
| 142 |
+
def update_dimensions_from_image(image_list):
|
| 143 |
+
"""Update width/height sliders based on uploaded image aspect ratio.
|
| 144 |
+
Keeps one side at 1024 and scales the other proportionally, with both sides as multiples of 8."""
|
| 145 |
+
if image_list is None or len(image_list) == 0:
|
| 146 |
+
return 1024, 1024 # Default dimensions
|
| 147 |
+
|
| 148 |
+
# Get the first image to determine dimensions
|
| 149 |
+
img = image_list[0][0] # Gallery returns list of tuples (image, caption)
|
| 150 |
+
img_width, img_height = img.size
|
| 151 |
+
|
| 152 |
+
aspect_ratio = img_width / img_height
|
| 153 |
+
|
| 154 |
+
if aspect_ratio >= 1: # Landscape or square
|
| 155 |
+
new_width = 1024
|
| 156 |
+
new_height = int(1024 / aspect_ratio)
|
| 157 |
+
else: # Portrait
|
| 158 |
+
new_height = 1024
|
| 159 |
+
new_width = int(1024 * aspect_ratio)
|
| 160 |
+
|
| 161 |
+
# Round to nearest multiple of 8
|
| 162 |
+
new_width = round(new_width / 8) * 8
|
| 163 |
+
new_height = round(new_height / 8) * 8
|
| 164 |
+
|
| 165 |
+
# Ensure within valid range (minimum 256, maximum 1024)
|
| 166 |
+
new_width = max(256, min(1024, new_width))
|
| 167 |
+
new_height = max(256, min(1024, new_height))
|
| 168 |
+
|
| 169 |
+
return new_width, new_height
|
| 170 |
+
|
| 171 |
# Updated duration function to match generate_image arguments (including progress)
|
| 172 |
+
def get_duration(prompt_embeds, image_list, width, height, num_inference_steps, guidance_scale, seed, progress=gr.Progress(track_tqdm=True)):
|
| 173 |
num_images = 0 if image_list is None else len(image_list)
|
| 174 |
step_duration = 1 + 0.8 * num_images
|
| 175 |
return max(65, num_inference_steps * step_duration + 10)
|
| 176 |
|
| 177 |
@spaces.GPU(duration=get_duration)
|
| 178 |
+
def generate_image(prompt_embeds, image_list, width, height, num_inference_steps, guidance_scale, seed, progress=gr.Progress(track_tqdm=True)):
|
| 179 |
# Move embeddings to GPU only when inside the GPU decorated function
|
| 180 |
prompt_embeds = prompt_embeds.to(device)
|
| 181 |
|
|
|
|
| 187 |
"num_inference_steps": num_inference_steps,
|
| 188 |
"guidance_scale": guidance_scale,
|
| 189 |
"generator": generator,
|
| 190 |
+
"width": width,
|
| 191 |
+
"height": height,
|
| 192 |
}
|
| 193 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 194 |
# Progress bar for the actual generation steps
|
| 195 |
if progress:
|
| 196 |
progress(0, desc="Starting generation...")
|
|
|
|
| 198 |
image = pipe(**pipe_kwargs).images[0]
|
| 199 |
return image
|
| 200 |
|
| 201 |
+
def infer(prompt, input_images=None, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=50, guidance_scale=2.5, prompt_upsampling=False, progress=gr.Progress(track_tqdm=True)):
|
| 202 |
|
| 203 |
if randomize_seed:
|
| 204 |
seed = random.randint(0, MAX_SEED)
|
|
|
|
| 233 |
num_inference_steps,
|
| 234 |
guidance_scale,
|
| 235 |
seed,
|
|
|
|
| 236 |
progress
|
| 237 |
)
|
| 238 |
|
|
|
|
| 253 |
css="""
|
| 254 |
#col-container {
|
| 255 |
margin: 0 auto;
|
| 256 |
+
max-width: 1200px;
|
| 257 |
}
|
| 258 |
.gallery-container img{
|
| 259 |
object-fit: contain;
|
|
|
|
| 266 |
gr.Markdown(f"""# FLUX.2 [dev]
|
| 267 |
FLUX.2 [dev] is a 32B model rectified flow capable of generating, editing and combining images based on text instructions model [[model](https://huggingface.co/black-forest-labs/FLUX.2-dev)], [[blog](https://bfl.ai/blog/flux-2)]
|
| 268 |
""")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 269 |
with gr.Row():
|
| 270 |
+
with gr.Column():
|
| 271 |
+
with gr.Row():
|
| 272 |
+
prompt = gr.Text(
|
| 273 |
+
label="Prompt",
|
| 274 |
+
show_label=False,
|
| 275 |
+
max_lines=2,
|
| 276 |
+
placeholder="Enter your prompt",
|
| 277 |
+
container=False,
|
| 278 |
+
scale=3
|
| 279 |
+
)
|
| 280 |
+
|
| 281 |
+
run_button = gr.Button("Run", scale=1)
|
| 282 |
+
|
| 283 |
+
with gr.Accordion("Input image(s) (optional)", open=True):
|
| 284 |
+
input_images = gr.Gallery(
|
| 285 |
+
label="Input Image(s)",
|
| 286 |
+
type="pil",
|
| 287 |
+
columns=3,
|
| 288 |
+
rows=1,
|
| 289 |
+
)
|
| 290 |
+
|
| 291 |
+
with gr.Accordion("Advanced Settings", open=False):
|
| 292 |
+
prompt_upsampling = gr.Checkbox(
|
| 293 |
+
label="Prompt Upsampling",
|
| 294 |
+
value=True,
|
| 295 |
+
info="Automatically enhance the prompt using a VLM"
|
| 296 |
+
)
|
| 297 |
|
| 298 |
+
seed = gr.Slider(
|
| 299 |
+
label="Seed",
|
| 300 |
+
minimum=0,
|
| 301 |
+
maximum=MAX_SEED,
|
| 302 |
+
step=1,
|
| 303 |
+
value=0,
|
| 304 |
+
)
|
| 305 |
+
|
| 306 |
+
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
| 307 |
+
|
| 308 |
+
with gr.Row():
|
| 309 |
+
|
| 310 |
+
width = gr.Slider(
|
| 311 |
+
label="Width",
|
| 312 |
+
minimum=256,
|
| 313 |
+
maximum=MAX_IMAGE_SIZE,
|
| 314 |
+
step=8,
|
| 315 |
+
value=1024,
|
| 316 |
+
)
|
| 317 |
+
|
| 318 |
+
height = gr.Slider(
|
| 319 |
+
label="Height",
|
| 320 |
+
minimum=256,
|
| 321 |
+
maximum=MAX_IMAGE_SIZE,
|
| 322 |
+
step=8,
|
| 323 |
+
value=1024,
|
| 324 |
+
)
|
| 325 |
+
|
| 326 |
+
with gr.Row():
|
| 327 |
+
|
| 328 |
+
num_inference_steps = gr.Slider(
|
| 329 |
+
label="Number of inference steps",
|
| 330 |
+
minimum=1,
|
| 331 |
+
maximum=100,
|
| 332 |
+
step=1,
|
| 333 |
+
value=30,
|
| 334 |
+
)
|
| 335 |
+
|
| 336 |
+
guidance_scale = gr.Slider(
|
| 337 |
+
label="Guidance scale",
|
| 338 |
+
minimum=0.0,
|
| 339 |
+
maximum=10.0,
|
| 340 |
+
step=0.1,
|
| 341 |
+
value=4,
|
| 342 |
+
)
|
| 343 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 344 |
|
| 345 |
+
with gr.Column():
|
| 346 |
+
result = gr.Image(label="Result", show_label=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 347 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 348 |
|
| 349 |
gr.Examples(
|
| 350 |
examples=examples,
|
|
|
|
| 364 |
cache_mode="lazy"
|
| 365 |
)
|
| 366 |
|
| 367 |
+
# Auto-update dimensions when images are uploaded
|
| 368 |
+
input_images.upload(
|
| 369 |
+
fn=update_dimensions_from_image,
|
| 370 |
+
inputs=[input_images],
|
| 371 |
+
outputs=[width, height]
|
| 372 |
+
)
|
| 373 |
+
|
| 374 |
gr.on(
|
| 375 |
triggers=[run_button.click, prompt.submit],
|
| 376 |
fn=infer,
|
| 377 |
+
inputs=[prompt, input_images, seed, randomize_seed, width, height, num_inference_steps, guidance_scale, prompt_upsampling],
|
| 378 |
outputs=[result, seed]
|
| 379 |
)
|
| 380 |
|