Spaces:
Running
on
Zero
Running
on
Zero
made use the same settings checkbox UI prettier
Browse files
app.py
CHANGED
@@ -84,16 +84,11 @@ def generate_single_image(
|
|
84 |
@spaces.GPU(duration=80)
|
85 |
def generate_arena_images(
|
86 |
prompt,
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
height_B,
|
93 |
-
width_A,
|
94 |
-
width_B,
|
95 |
-
guidance_scale_A,
|
96 |
-
guidance_scale_B,
|
97 |
seed,
|
98 |
num_images_per_prompt,
|
99 |
model_choice_A,
|
@@ -106,22 +101,14 @@ def generate_arena_images(
|
|
106 |
|
107 |
generator = torch.Generator().manual_seed(seed)
|
108 |
|
109 |
-
# Apply settings based on use_same_settings
|
110 |
-
if use_same_settings:
|
111 |
-
num_inference_steps_B = num_inference_steps_A
|
112 |
-
height_B = height_A
|
113 |
-
width_B = width_A
|
114 |
-
guidance_scale_B = guidance_scale_A
|
115 |
-
negative_prompt_B = negative_prompt_A
|
116 |
-
|
117 |
# Generate images for both models
|
118 |
images_A = generate_single_image(
|
119 |
prompt,
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
seed,
|
126 |
num_images_per_prompt,
|
127 |
model_choice_A,
|
@@ -129,11 +116,11 @@ def generate_arena_images(
|
|
129 |
)
|
130 |
images_B = generate_single_image(
|
131 |
prompt,
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
seed,
|
138 |
num_images_per_prompt,
|
139 |
model_choice_B,
|
@@ -228,7 +215,69 @@ with gr.Blocks(css=css) as demo:
|
|
228 |
result_B = gr.Gallery(label="Generated Images (Model B)", elem_id="gallery_B")
|
229 |
with gr.Accordion("Advanced options", open=False):
|
230 |
use_same_settings = gr.Checkbox(label='Use same settings for both models', value=True)
|
231 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
232 |
negative_prompt_A = gr.Textbox(
|
233 |
label="Negative Prompt (Model A)",
|
234 |
info="Describe what you don't want in the image",
|
@@ -241,7 +290,7 @@ with gr.Blocks(css=css) as demo:
|
|
241 |
value="deformed, distorted, disfigured, poorly drawn, bad anatomy, incorrect anatomy, extra limb, missing limb, floating limbs, mutated hands and fingers, disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation",
|
242 |
placeholder="Ugly, bad anatomy...",
|
243 |
)
|
244 |
-
with gr.Row():
|
245 |
num_inference_steps_A = gr.Slider(
|
246 |
label="Number of Inference Steps (Model A)",
|
247 |
info="The number of denoising steps of the image. More denoising steps usually lead to a higher quality image at the cost of slower inference",
|
@@ -258,7 +307,7 @@ with gr.Blocks(css=css) as demo:
|
|
258 |
value=25,
|
259 |
step=1,
|
260 |
)
|
261 |
-
with gr.Row():
|
262 |
width_A = gr.Slider(
|
263 |
label="Width (Model A)",
|
264 |
info="Width of the Image",
|
@@ -275,7 +324,7 @@ with gr.Blocks(css=css) as demo:
|
|
275 |
step=32,
|
276 |
value=1024,
|
277 |
)
|
278 |
-
with gr.Row():
|
279 |
height_A = gr.Slider(
|
280 |
label="Height (Model A)",
|
281 |
info="Height of the Image",
|
@@ -292,7 +341,7 @@ with gr.Blocks(css=css) as demo:
|
|
292 |
step=32,
|
293 |
value=1024,
|
294 |
)
|
295 |
-
with gr.Row():
|
296 |
guidance_scale_A = gr.Slider(
|
297 |
label="Guidance Scale (Model A)",
|
298 |
info="Controls how much the image generation process follows the text prompt. Higher values make the image stick more closely to the input text.",
|
@@ -309,17 +358,34 @@ with gr.Blocks(css=css) as demo:
|
|
309 |
value=7.5,
|
310 |
step=0.1,
|
311 |
)
|
312 |
-
with gr.Row():
|
313 |
-
|
314 |
value=42,
|
315 |
minimum=0,
|
316 |
maximum=MAX_SEED,
|
317 |
step=1,
|
318 |
-
label="Seed",
|
319 |
info="A starting point to initiate the generation process, put 0 for a random one",
|
320 |
)
|
321 |
-
|
322 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
323 |
info="Number of Images to generate with the settings",
|
324 |
minimum=1,
|
325 |
maximum=4,
|
@@ -342,21 +408,16 @@ with gr.Blocks(css=css) as demo:
|
|
342 |
fn=generate_arena_images,
|
343 |
inputs=[
|
344 |
prompt,
|
345 |
-
|
346 |
-
|
347 |
-
|
348 |
-
|
349 |
-
|
350 |
-
height_B,
|
351 |
-
width_A,
|
352 |
-
width_B,
|
353 |
-
guidance_scale_A,
|
354 |
-
guidance_scale_B,
|
355 |
seed,
|
356 |
num_images_per_prompt,
|
357 |
model_choice_A,
|
358 |
model_choice_B,
|
359 |
-
use_same_settings
|
360 |
],
|
361 |
outputs=[result_A, result_B],
|
362 |
)
|
|
|
84 |
@spaces.GPU(duration=80)
|
85 |
def generate_arena_images(
|
86 |
prompt,
|
87 |
+
negative_prompt,
|
88 |
+
num_inference_steps,
|
89 |
+
height,
|
90 |
+
width,
|
91 |
+
guidance_scale,
|
|
|
|
|
|
|
|
|
|
|
92 |
seed,
|
93 |
num_images_per_prompt,
|
94 |
model_choice_A,
|
|
|
101 |
|
102 |
generator = torch.Generator().manual_seed(seed)
|
103 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
# Generate images for both models
|
105 |
images_A = generate_single_image(
|
106 |
prompt,
|
107 |
+
negative_prompt,
|
108 |
+
num_inference_steps,
|
109 |
+
height,
|
110 |
+
width,
|
111 |
+
guidance_scale,
|
112 |
seed,
|
113 |
num_images_per_prompt,
|
114 |
model_choice_A,
|
|
|
116 |
)
|
117 |
images_B = generate_single_image(
|
118 |
prompt,
|
119 |
+
negative_prompt,
|
120 |
+
num_inference_steps,
|
121 |
+
height,
|
122 |
+
width,
|
123 |
+
guidance_scale,
|
124 |
seed,
|
125 |
num_images_per_prompt,
|
126 |
model_choice_B,
|
|
|
215 |
result_B = gr.Gallery(label="Generated Images (Model B)", elem_id="gallery_B")
|
216 |
with gr.Accordion("Advanced options", open=False):
|
217 |
use_same_settings = gr.Checkbox(label='Use same settings for both models', value=True)
|
218 |
+
|
219 |
+
# Conditional UI elements based on use_same_settings
|
220 |
+
with gr.Row(visible=True):
|
221 |
+
negative_prompt = gr.Textbox(
|
222 |
+
label="Negative Prompt",
|
223 |
+
info="Describe what you don't want in the image",
|
224 |
+
value="deformed, distorted, disfigured, poorly drawn, bad anatomy, incorrect anatomy, extra limb, missing limb, floating limbs, mutated hands and fingers, disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation",
|
225 |
+
placeholder="Ugly, bad anatomy...",
|
226 |
+
)
|
227 |
+
with gr.Row(visible=True):
|
228 |
+
num_inference_steps = gr.Slider(
|
229 |
+
label="Number of Inference Steps",
|
230 |
+
info="The number of denoising steps of the image. More denoising steps usually lead to a higher quality image at the cost of slower inference",
|
231 |
+
minimum=1,
|
232 |
+
maximum=50,
|
233 |
+
value=25,
|
234 |
+
step=1,
|
235 |
+
)
|
236 |
+
guidance_scale = gr.Slider(
|
237 |
+
label="Guidance Scale",
|
238 |
+
info="Controls how much the image generation process follows the text prompt. Higher values make the image stick more closely to the input text.",
|
239 |
+
minimum=0.0,
|
240 |
+
maximum=10.0,
|
241 |
+
value=7.5,
|
242 |
+
step=0.1,
|
243 |
+
)
|
244 |
+
with gr.Row(visible=True):
|
245 |
+
width = gr.Slider(
|
246 |
+
label="Width",
|
247 |
+
info="Width of the Image",
|
248 |
+
minimum=256,
|
249 |
+
maximum=1344,
|
250 |
+
step=32,
|
251 |
+
value=1024,
|
252 |
+
)
|
253 |
+
height = gr.Slider(
|
254 |
+
label="Height",
|
255 |
+
info="Height of the Image",
|
256 |
+
minimum=256,
|
257 |
+
maximum=1344,
|
258 |
+
step=32,
|
259 |
+
value=1024,
|
260 |
+
)
|
261 |
+
with gr.Row(visible=True):
|
262 |
+
seed = gr.Slider(
|
263 |
+
value=42,
|
264 |
+
minimum=0,
|
265 |
+
maximum=MAX_SEED,
|
266 |
+
step=1,
|
267 |
+
label="Seed",
|
268 |
+
info="A starting point to initiate the generation process, put 0 for a random one",
|
269 |
+
)
|
270 |
+
num_images_per_prompt = gr.Slider(
|
271 |
+
label="Images Per Prompt",
|
272 |
+
info="Number of Images to generate with the settings",
|
273 |
+
minimum=1,
|
274 |
+
maximum=4,
|
275 |
+
step=1,
|
276 |
+
value=2,
|
277 |
+
)
|
278 |
+
|
279 |
+
# Conditional UI elements based on use_same_settings
|
280 |
+
with gr.Row(visible=False):
|
281 |
negative_prompt_A = gr.Textbox(
|
282 |
label="Negative Prompt (Model A)",
|
283 |
info="Describe what you don't want in the image",
|
|
|
290 |
value="deformed, distorted, disfigured, poorly drawn, bad anatomy, incorrect anatomy, extra limb, missing limb, floating limbs, mutated hands and fingers, disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation",
|
291 |
placeholder="Ugly, bad anatomy...",
|
292 |
)
|
293 |
+
with gr.Row(visible=False):
|
294 |
num_inference_steps_A = gr.Slider(
|
295 |
label="Number of Inference Steps (Model A)",
|
296 |
info="The number of denoising steps of the image. More denoising steps usually lead to a higher quality image at the cost of slower inference",
|
|
|
307 |
value=25,
|
308 |
step=1,
|
309 |
)
|
310 |
+
with gr.Row(visible=False):
|
311 |
width_A = gr.Slider(
|
312 |
label="Width (Model A)",
|
313 |
info="Width of the Image",
|
|
|
324 |
step=32,
|
325 |
value=1024,
|
326 |
)
|
327 |
+
with gr.Row(visible=False):
|
328 |
height_A = gr.Slider(
|
329 |
label="Height (Model A)",
|
330 |
info="Height of the Image",
|
|
|
341 |
step=32,
|
342 |
value=1024,
|
343 |
)
|
344 |
+
with gr.Row(visible=False):
|
345 |
guidance_scale_A = gr.Slider(
|
346 |
label="Guidance Scale (Model A)",
|
347 |
info="Controls how much the image generation process follows the text prompt. Higher values make the image stick more closely to the input text.",
|
|
|
358 |
value=7.5,
|
359 |
step=0.1,
|
360 |
)
|
361 |
+
with gr.Row(visible=False):
|
362 |
+
seed_A = gr.Slider(
|
363 |
value=42,
|
364 |
minimum=0,
|
365 |
maximum=MAX_SEED,
|
366 |
step=1,
|
367 |
+
label="Seed (Model A)",
|
368 |
info="A starting point to initiate the generation process, put 0 for a random one",
|
369 |
)
|
370 |
+
seed_B = gr.Slider(
|
371 |
+
value=42,
|
372 |
+
minimum=0,
|
373 |
+
maximum=MAX_SEED,
|
374 |
+
step=1,
|
375 |
+
label="Seed (Model B)",
|
376 |
+
info="A starting point to initiate the generation process, put 0 for a random one",
|
377 |
+
)
|
378 |
+
with gr.Row(visible=False):
|
379 |
+
num_images_per_prompt_A = gr.Slider(
|
380 |
+
label="Images Per Prompt (Model A)",
|
381 |
+
info="Number of Images to generate with the settings",
|
382 |
+
minimum=1,
|
383 |
+
maximum=4,
|
384 |
+
step=1,
|
385 |
+
value=2,
|
386 |
+
)
|
387 |
+
num_images_per_prompt_B = gr.Slider(
|
388 |
+
label="Images Per Prompt (Model B)",
|
389 |
info="Number of Images to generate with the settings",
|
390 |
minimum=1,
|
391 |
maximum=4,
|
|
|
408 |
fn=generate_arena_images,
|
409 |
inputs=[
|
410 |
prompt,
|
411 |
+
negative_prompt,
|
412 |
+
num_inference_steps,
|
413 |
+
height,
|
414 |
+
width,
|
415 |
+
guidance_scale,
|
|
|
|
|
|
|
|
|
|
|
416 |
seed,
|
417 |
num_images_per_prompt,
|
418 |
model_choice_A,
|
419 |
model_choice_B,
|
420 |
+
use_same_settings,
|
421 |
],
|
422 |
outputs=[result_A, result_B],
|
423 |
)
|