Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -11,56 +11,52 @@ from pipeline_fill_sd_xl import StableDiffusionXLFillPipeline
|
|
11 |
from PIL import Image, ImageDraw
|
12 |
import numpy as np
|
13 |
|
14 |
-
#
|
15 |
-
|
16 |
-
# Load ControlNet Union
|
17 |
config_file = hf_hub_download(
|
18 |
"xinsir/controlnet-union-sdxl-1.0",
|
19 |
filename="config_promax.json",
|
20 |
)
|
|
|
21 |
config = ControlNetModel_Union.load_config(config_file)
|
22 |
controlnet_model = ControlNetModel_Union.from_config(config)
|
23 |
model_file = hf_hub_download(
|
24 |
"xinsir/controlnet-union-sdxl-1.0",
|
25 |
filename="diffusion_pytorch_model_promax.safetensors",
|
26 |
)
|
|
|
27 |
sstate_dict = load_state_dict(model_file)
|
28 |
model, _, _, _, _ = ControlNetModel_Union._load_pretrained_model(
|
29 |
controlnet_model, sstate_dict, model_file, "xinsir/controlnet-union-sdxl-1.0"
|
30 |
)
|
31 |
model.to(device="cuda", dtype=torch.float16)
|
32 |
|
33 |
-
# Load VAE
|
34 |
vae = AutoencoderKL.from_pretrained(
|
35 |
"madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16
|
36 |
).to("cuda")
|
37 |
|
38 |
-
#
|
39 |
-
|
40 |
-
|
41 |
-
# Load RealVisXL V5.0 Lightning
|
42 |
-
pipe_v5 = StableDiffusionXLFillPipeline.from_pretrained(
|
43 |
"SG161222/RealVisXL_V5.0_Lightning",
|
44 |
torch_dtype=torch.float16,
|
45 |
vae=vae,
|
46 |
-
controlnet=model,
|
47 |
-
variant="fp16",
|
48 |
-
).to("cuda")
|
49 |
-
pipe_v5.scheduler = TCDScheduler.from_config(pipe_v5.scheduler.config)
|
50 |
-
pipelines["RealVisXL V5.0 Lightning"] = pipe_v5
|
51 |
-
|
52 |
-
# Load RealVisXL V4.0 Lightning
|
53 |
-
pipe_v4 = StableDiffusionXLFillPipeline.from_pretrained(
|
54 |
-
"SG161222/RealVisXL_V4.0_Lightning",
|
55 |
-
torch_dtype=torch.float16,
|
56 |
-
vae=vae, # Use the same VAE
|
57 |
-
controlnet=model, # Use the same controlnet
|
58 |
variant="fp16",
|
59 |
).to("cuda")
|
60 |
-
pipe_v4.scheduler = TCDScheduler.from_config(pipe_v4.scheduler.config)
|
61 |
-
pipelines["RealVisXL V4.0 Lightning"] = pipe_v4
|
62 |
|
63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
65 |
def prepare_image_and_mask(image, width, height, overlap_percentage, resize_option, custom_resize_percentage, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom):
|
66 |
target_size = (width, height)
|
@@ -69,7 +65,7 @@ def prepare_image_and_mask(image, width, height, overlap_percentage, resize_opti
|
|
69 |
scale_factor = min(target_size[0] / image.width, target_size[1] / image.height)
|
70 |
new_width = int(image.width * scale_factor)
|
71 |
new_height = int(image.height * scale_factor)
|
72 |
-
|
73 |
# Resize the source image to fit within target size
|
74 |
source = image.resize((new_width, new_height), Image.LANCZOS)
|
75 |
|
@@ -121,10 +117,6 @@ def prepare_image_and_mask(image, width, height, overlap_percentage, resize_opti
|
|
121 |
elif alignment == "Bottom":
|
122 |
margin_x = (target_size[0] - new_width) // 2
|
123 |
margin_y = target_size[1] - new_height
|
124 |
-
else: # Default to Middle if alignment is somehow invalid
|
125 |
-
margin_x = (target_size[0] - new_width) // 2
|
126 |
-
margin_y = (target_size[1] - new_height) // 2
|
127 |
-
|
128 |
|
129 |
# Adjust margins to eliminate gaps
|
130 |
margin_x = max(0, min(margin_x, target_size[0] - new_width))
|
@@ -135,126 +127,66 @@ def prepare_image_and_mask(image, width, height, overlap_percentage, resize_opti
|
|
135 |
background.paste(source, (margin_x, margin_y))
|
136 |
|
137 |
# Create the mask
|
138 |
-
mask = Image.new('L', target_size, 255)
|
139 |
mask_draw = ImageDraw.Draw(mask)
|
140 |
|
141 |
-
# Calculate overlap areas
|
142 |
-
white_gaps_patch = 2
|
143 |
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
if overlap_right:
|
159 |
-
right_black -= overlap_x
|
160 |
-
else:
|
161 |
-
# If not overlapping right, ensure the black mask ends exactly at the image edge or slightly inside
|
162 |
-
right_black -= white_gaps_patch if alignment != "Right" else 0
|
163 |
-
|
164 |
-
if overlap_top:
|
165 |
-
top_black += overlap_y
|
166 |
-
else:
|
167 |
-
# If not overlapping top, ensure the black mask starts exactly at the image edge or slightly inside
|
168 |
-
top_black += white_gaps_patch if alignment != "Top" else 0
|
169 |
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
# Ensure coordinates are valid (left < right, top < bottom)
|
177 |
-
left_black = min(left_black, target_size[0])
|
178 |
-
top_black = min(top_black, target_size[1])
|
179 |
-
right_black = max(left_black, right_black) # Ensure right >= left
|
180 |
-
bottom_black = max(top_black, bottom_black) # Ensure bottom >= top
|
181 |
-
right_black = min(right_black, target_size[0])
|
182 |
-
bottom_black = min(bottom_black, target_size[1])
|
183 |
-
|
184 |
-
|
185 |
-
# Draw the black rectangle onto the white mask
|
186 |
-
# The area *inside* this rectangle will be kept (mask value 0)
|
187 |
-
# The area *outside* this rectangle will be filled (mask value 255)
|
188 |
-
if right_black > left_black and bottom_black > top_black:
|
189 |
-
mask_draw.rectangle(
|
190 |
-
[(left_black, top_black), (right_black, bottom_black)],
|
191 |
-
fill=0 # Black means keep this area
|
192 |
-
)
|
193 |
|
194 |
return background, mask
|
195 |
|
196 |
-
|
197 |
@spaces.GPU(duration=24)
|
198 |
-
def infer(
|
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 |
-
result_image = pipe(
|
230 |
-
prompt_embeds=prompt_embeds,
|
231 |
-
negative_prompt_embeds=negative_prompt_embeds,
|
232 |
-
pooled_prompt_embeds=pooled_prompt_embeds,
|
233 |
-
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
|
234 |
-
image=background, # The background containing the original image
|
235 |
-
mask_image=mask, # The mask (white = fill, black = keep)
|
236 |
-
control_image=cnet_image, # ControlNet input image
|
237 |
-
num_inference_steps=num_inference_steps,
|
238 |
-
generator=generator, # Use generator for reproducibility if needed
|
239 |
-
output_type="pil" # Ensure PIL output
|
240 |
-
).images[0]
|
241 |
-
|
242 |
-
# The pipeline directly returns the final composited image.
|
243 |
-
# No need for manual pasting like before.
|
244 |
-
|
245 |
-
return result_image
|
246 |
-
except Exception as e:
|
247 |
-
print(f"Error during inference: {e}")
|
248 |
-
import traceback
|
249 |
-
traceback.print_exc()
|
250 |
-
# Return the background image or raise a Gradio error for clarity
|
251 |
-
# raise gr.Error(f"Inference failed: {e}")
|
252 |
-
# Or return the prepared background/mask for debugging
|
253 |
-
background, mask = prepare_image_and_mask(image, width, height, overlap_percentage, resize_option, custom_resize_percentage, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom)
|
254 |
-
# Combine background and mask for visualization
|
255 |
-
debug_img = Image.blend(background.convert("RGBA"), mask.convert("RGBA"), 0.5)
|
256 |
-
return debug_img # Return a debug image or None
|
257 |
-
|
258 |
|
259 |
def clear_result():
|
260 |
"""Clears the result Image."""
|
@@ -265,21 +197,17 @@ def preload_presets(target_ratio, ui_width, ui_height):
|
|
265 |
if target_ratio == "9:16":
|
266 |
changed_width = 720
|
267 |
changed_height = 1280
|
268 |
-
return changed_width, changed_height, gr.update(
|
269 |
elif target_ratio == "16:9":
|
270 |
changed_width = 1280
|
271 |
changed_height = 720
|
272 |
-
return changed_width, changed_height, gr.update(
|
273 |
elif target_ratio == "1:1":
|
274 |
changed_width = 1024
|
275 |
changed_height = 1024
|
276 |
-
return changed_width, changed_height, gr.update(
|
277 |
elif target_ratio == "Custom":
|
278 |
-
# When switching to Custom, keep current slider values but open accordion
|
279 |
return ui_width, ui_height, gr.update(open=True)
|
280 |
-
# Should not happen, but return current values if it does
|
281 |
-
return ui_width, ui_height, gr.update()
|
282 |
-
|
283 |
|
284 |
def select_the_right_preset(user_width, user_height):
|
285 |
if user_width == 720 and user_height == 1280:
|
@@ -296,71 +224,54 @@ def toggle_custom_resize_slider(resize_option):
|
|
296 |
|
297 |
def update_history(new_image, history):
|
298 |
"""Updates the history gallery with the new image."""
|
299 |
-
if new_image is None: # Don't add None to history (e.g., on clear or error)
|
300 |
-
return history
|
301 |
if history is None:
|
302 |
history = []
|
303 |
-
# Prepend the new image (as PIL or path depending on Gallery config)
|
304 |
history.insert(0, new_image)
|
305 |
-
# Limit history size if desired (e.g., keep last 12)
|
306 |
-
max_history = 12
|
307 |
-
if len(history) > max_history:
|
308 |
-
history = history[:max_history]
|
309 |
return history
|
310 |
|
311 |
-
#
|
312 |
css = """
|
313 |
h1 {
|
314 |
-
|
315 |
-
|
316 |
-
}
|
317 |
-
.gradio-container {
|
318 |
-
max-width: 1280px !important;
|
319 |
-
margin: auto !important;
|
320 |
}
|
321 |
"""
|
322 |
|
323 |
title = """<h1 align="center">Diffusers Image Outpaint Lightning</h1>
|
324 |
-
<p align="center">Expand images using ControlNet Union and Lightning models. Choose a base model below.</p>
|
325 |
"""
|
326 |
|
327 |
-
# --- Gradio UI ---
|
328 |
with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
|
329 |
with gr.Column():
|
330 |
gr.HTML(title)
|
331 |
|
332 |
with gr.Row():
|
333 |
-
with gr.Column(
|
334 |
input_image = gr.Image(
|
335 |
type="pil",
|
336 |
label="Input Image"
|
337 |
)
|
338 |
-
|
339 |
-
|
340 |
-
|
341 |
-
label="Select Model"
|
342 |
-
choices=list(pipelines.keys()),
|
343 |
-
value="RealVisXL V5.0 Lightning", # Default model
|
344 |
)
|
345 |
-
|
346 |
with gr.Row():
|
347 |
with gr.Column(scale=2):
|
348 |
-
prompt_input = gr.Textbox(label="Prompt (
|
349 |
-
with gr.Column(scale=1
|
350 |
-
run_button = gr.Button("Generate"
|
351 |
|
352 |
with gr.Row():
|
353 |
target_ratio = gr.Radio(
|
354 |
-
label="
|
355 |
choices=["9:16", "16:9", "1:1", "Custom"],
|
356 |
-
value="9:16",
|
357 |
scale=2
|
358 |
)
|
359 |
-
|
360 |
alignment_dropdown = gr.Dropdown(
|
361 |
choices=["Middle", "Left", "Right", "Top", "Bottom"],
|
362 |
value="Middle",
|
363 |
-
label="
|
364 |
)
|
365 |
|
366 |
with gr.Accordion(label="Advanced settings", open=False) as settings_panel:
|
@@ -368,43 +279,38 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
|
|
368 |
with gr.Row():
|
369 |
width_slider = gr.Slider(
|
370 |
label="Target Width",
|
371 |
-
minimum=
|
372 |
maximum=1536,
|
373 |
-
step=
|
374 |
-
value=720,
|
375 |
)
|
376 |
height_slider = gr.Slider(
|
377 |
label="Target Height",
|
378 |
-
minimum=
|
379 |
maximum=1536,
|
380 |
-
step=
|
381 |
-
value=1280,
|
382 |
)
|
383 |
-
|
384 |
num_inference_steps = gr.Slider(label="Steps", minimum=4, maximum=12, step=1, value=8)
|
385 |
-
|
386 |
with gr.Group():
|
387 |
overlap_percentage = gr.Slider(
|
388 |
label="Mask overlap (%)",
|
389 |
-
info="Percentage of the input image edge to keep (reduces seams)",
|
390 |
minimum=1,
|
391 |
maximum=50,
|
392 |
-
value=10,
|
393 |
step=1
|
394 |
)
|
395 |
-
gr.Markdown("Select edges to apply overlap:")
|
396 |
with gr.Row():
|
397 |
-
overlap_top = gr.Checkbox(label="Top", value=True)
|
398 |
-
overlap_right = gr.Checkbox(label="Right", value=True)
|
399 |
-
|
400 |
-
|
401 |
-
|
402 |
with gr.Row():
|
403 |
resize_option = gr.Radio(
|
404 |
-
label="Resize input image
|
405 |
-
info="Scale the input image relative to its fitted size",
|
406 |
choices=["Full", "50%", "33%", "25%", "Custom"],
|
407 |
-
value="Full"
|
408 |
)
|
409 |
custom_resize_percentage = gr.Slider(
|
410 |
label="Custom resize (%)",
|
@@ -412,48 +318,35 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
|
|
412 |
maximum=100,
|
413 |
step=1,
|
414 |
value=50,
|
415 |
-
visible=False
|
416 |
)
|
417 |
-
|
418 |
gr.Examples(
|
419 |
examples=[
|
420 |
-
["./examples/example_1.webp",
|
421 |
-
["./examples/example_2.jpg",
|
422 |
-
["./examples/example_3.jpg",
|
423 |
-
["./examples/example_3.jpg",
|
424 |
],
|
425 |
-
inputs=[input_image,
|
426 |
-
label="Examples (Prompt is optional)"
|
427 |
)
|
428 |
|
429 |
-
with gr.Column(
|
430 |
result = gr.Image(
|
431 |
interactive=False,
|
432 |
label="Generated Image",
|
433 |
format="png",
|
434 |
)
|
435 |
-
history_gallery = gr.Gallery(
|
436 |
-
label="History",
|
437 |
-
columns=4, # Adjust columns as needed
|
438 |
-
object_fit="contain",
|
439 |
-
interactive=False,
|
440 |
-
show_label=True,
|
441 |
-
allow_preview=True,
|
442 |
-
preview=True
|
443 |
-
)
|
444 |
-
|
445 |
-
|
446 |
-
# --- Event Listeners ---
|
447 |
|
448 |
-
#
|
|
|
449 |
target_ratio.change(
|
450 |
fn=preload_presets,
|
451 |
inputs=[target_ratio, width_slider, height_slider],
|
452 |
outputs=[width_slider, height_slider, settings_panel],
|
453 |
queue=False
|
454 |
)
|
455 |
-
|
456 |
-
# Update ratio selection based on slider changes
|
457 |
width_slider.change(
|
458 |
fn=select_the_right_preset,
|
459 |
inputs=[width_slider, height_slider],
|
@@ -466,55 +359,42 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
|
|
466 |
outputs=[target_ratio],
|
467 |
queue=False
|
468 |
)
|
469 |
-
|
470 |
-
# Show/hide custom resize slider
|
471 |
resize_option.change(
|
472 |
fn=toggle_custom_resize_slider,
|
473 |
inputs=[resize_option],
|
474 |
outputs=[custom_resize_percentage],
|
475 |
queue=False
|
476 |
)
|
477 |
-
|
478 |
-
# Define inputs for the main inference function
|
479 |
-
infer_inputs = [
|
480 |
-
model_selector, input_image, width_slider, height_slider, overlap_percentage,
|
481 |
-
num_inference_steps, resize_option, custom_resize_percentage, prompt_input,
|
482 |
-
alignment_dropdown, overlap_left, overlap_right, overlap_top, overlap_bottom
|
483 |
-
]
|
484 |
-
|
485 |
-
# --- Run Button Click ---
|
486 |
run_button.click(
|
487 |
fn=clear_result,
|
488 |
inputs=None,
|
489 |
-
outputs=
|
490 |
-
queue=False # Clearing should be fast
|
491 |
).then(
|
492 |
fn=infer,
|
493 |
-
inputs=
|
494 |
-
|
|
|
|
|
495 |
).then(
|
496 |
-
fn=
|
497 |
-
inputs=[result, history_gallery],
|
498 |
-
outputs=
|
499 |
)
|
500 |
-
|
501 |
-
# --- Prompt Submit (Enter Key) ---
|
502 |
prompt_input.submit(
|
503 |
-
|
504 |
inputs=None,
|
505 |
-
outputs=
|
506 |
-
queue=False
|
507 |
).then(
|
508 |
fn=infer,
|
509 |
-
inputs=
|
510 |
-
|
|
|
|
|
511 |
).then(
|
512 |
-
fn=update_history,
|
513 |
inputs=[result, history_gallery],
|
514 |
-
outputs=
|
515 |
)
|
|
|
516 |
|
517 |
-
# --- Launch App ---
|
518 |
-
# Make sure you have example images at the specified paths or remove/update the gr.Examples section
|
519 |
-
# Create an 'examples' directory and place images like 'example_1.webp', 'example_2.jpg', 'example_3.jpg' inside it.
|
520 |
demo.queue(max_size=20).launch(share=False, ssr_mode=False, show_error=True)
|
|
|
11 |
from PIL import Image, ImageDraw
|
12 |
import numpy as np
|
13 |
|
14 |
+
# Load configuration and models
|
|
|
|
|
15 |
config_file = hf_hub_download(
|
16 |
"xinsir/controlnet-union-sdxl-1.0",
|
17 |
filename="config_promax.json",
|
18 |
)
|
19 |
+
|
20 |
config = ControlNetModel_Union.load_config(config_file)
|
21 |
controlnet_model = ControlNetModel_Union.from_config(config)
|
22 |
model_file = hf_hub_download(
|
23 |
"xinsir/controlnet-union-sdxl-1.0",
|
24 |
filename="diffusion_pytorch_model_promax.safetensors",
|
25 |
)
|
26 |
+
|
27 |
sstate_dict = load_state_dict(model_file)
|
28 |
model, _, _, _, _ = ControlNetModel_Union._load_pretrained_model(
|
29 |
controlnet_model, sstate_dict, model_file, "xinsir/controlnet-union-sdxl-1.0"
|
30 |
)
|
31 |
model.to(device="cuda", dtype=torch.float16)
|
32 |
|
|
|
33 |
vae = AutoencoderKL.from_pretrained(
|
34 |
"madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16
|
35 |
).to("cuda")
|
36 |
|
37 |
+
# Initially load the default pipeline
|
38 |
+
pipe = StableDiffusionXLFillPipeline.from_pretrained(
|
|
|
|
|
|
|
39 |
"SG161222/RealVisXL_V5.0_Lightning",
|
40 |
torch_dtype=torch.float16,
|
41 |
vae=vae,
|
42 |
+
controlnet=model,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
variant="fp16",
|
44 |
).to("cuda")
|
|
|
|
|
45 |
|
46 |
+
pipe.scheduler = TCDScheduler.from_config(pipe.scheduler.config)
|
47 |
+
|
48 |
+
def load_model(selected_model):
|
49 |
+
global pipe
|
50 |
+
model_path = f"SG161222/{selected_model}"
|
51 |
+
pipe = StableDiffusionXLFillPipeline.from_pretrained(
|
52 |
+
model_path,
|
53 |
+
torch_dtype=torch.float16,
|
54 |
+
vae=vae,
|
55 |
+
controlnet=model,
|
56 |
+
variant="fp16",
|
57 |
+
).to("cuda")
|
58 |
+
pipe.scheduler = TCDScheduler.from_config(pipe.scheduler.config)
|
59 |
+
return f"Loaded model: {selected_model}"
|
60 |
|
61 |
def prepare_image_and_mask(image, width, height, overlap_percentage, resize_option, custom_resize_percentage, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom):
|
62 |
target_size = (width, height)
|
|
|
65 |
scale_factor = min(target_size[0] / image.width, target_size[1] / image.height)
|
66 |
new_width = int(image.width * scale_factor)
|
67 |
new_height = int(image.height * scale_factor)
|
68 |
+
|
69 |
# Resize the source image to fit within target size
|
70 |
source = image.resize((new_width, new_height), Image.LANCZOS)
|
71 |
|
|
|
117 |
elif alignment == "Bottom":
|
118 |
margin_x = (target_size[0] - new_width) // 2
|
119 |
margin_y = target_size[1] - new_height
|
|
|
|
|
|
|
|
|
120 |
|
121 |
# Adjust margins to eliminate gaps
|
122 |
margin_x = max(0, min(margin_x, target_size[0] - new_width))
|
|
|
127 |
background.paste(source, (margin_x, margin_y))
|
128 |
|
129 |
# Create the mask
|
130 |
+
mask = Image.new('L', target_size, 255)
|
131 |
mask_draw = ImageDraw.Draw(mask)
|
132 |
|
133 |
+
# Calculate overlap areas
|
134 |
+
white_gaps_patch = 2
|
135 |
|
136 |
+
left_overlap = margin_x + overlap_x if overlap_left else margin_x + white_gaps_patch
|
137 |
+
right_overlap = margin_x + new_width - overlap_x if overlap_right else margin_x + new_width - white_gaps_patch
|
138 |
+
top_overlap = margin_y + overlap_y if overlap_top else margin_y + white_gaps_patch
|
139 |
+
bottom_overlap = margin_y + new_height - overlap_y if overlap_bottom else margin_y + new_height - white_gaps_patch
|
140 |
+
|
141 |
+
if alignment == "Left":
|
142 |
+
left_overlap = margin_x + overlap_x if overlap_left else margin_x
|
143 |
+
elif alignment == "Right":
|
144 |
+
right_overlap = margin_x + new_width - overlap_x if overlap_right else margin_x + new_width
|
145 |
+
elif alignment == "Top":
|
146 |
+
top_overlap = margin_y + overlap_y if overlap_top else margin_y
|
147 |
+
elif alignment == "Bottom":
|
148 |
+
bottom_overlap = margin_y + new_height - overlap_y if overlap_bottom else margin_y + new_height
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
149 |
|
150 |
+
# Draw the mask
|
151 |
+
mask_draw.rectangle([
|
152 |
+
(left_overlap, top_overlap),
|
153 |
+
(right_overlap, bottom_overlap)
|
154 |
+
], fill=0)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
155 |
|
156 |
return background, mask
|
157 |
|
|
|
158 |
@spaces.GPU(duration=24)
|
159 |
+
def infer(image, width, height, overlap_percentage, num_inference_steps, resize_option, custom_resize_percentage, prompt_input, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom):
|
160 |
+
background, mask = prepare_image_and_mask(image, width, height, overlap_percentage, resize_option, custom_resize_percentage, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom)
|
161 |
+
|
162 |
+
cnet_image = background.copy()
|
163 |
+
cnet_image.paste(0, (0, 0), mask)
|
164 |
+
|
165 |
+
final_prompt = f"{prompt_input} , high quality, 4k"
|
166 |
+
|
167 |
+
(
|
168 |
+
prompt_embeds,
|
169 |
+
negative_prompt_embeds,
|
170 |
+
pooled_prompt_embeds,
|
171 |
+
negative_pooled_prompt_embeds,
|
172 |
+
) = pipe.encode_prompt(final_prompt, "cuda", True)
|
173 |
+
|
174 |
+
# Generate the image
|
175 |
+
for image in pipe(
|
176 |
+
prompt_embeds=prompt_embeds,
|
177 |
+
negative_prompt_embeds=negative_prompt_embeds,
|
178 |
+
pooled_prompt_embeds=pooled_prompt_embeds,
|
179 |
+
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
|
180 |
+
image=cnet_image,
|
181 |
+
num_inference_steps=num_inference_steps
|
182 |
+
):
|
183 |
+
pass # Wait for the generation to complete
|
184 |
+
generated_image = image # Get the last image
|
185 |
+
|
186 |
+
generated_image = generated_image.convert("RGBA")
|
187 |
+
cnet_image.paste(generated_image, (0, 0), mask)
|
188 |
+
|
189 |
+
return cnet_image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
190 |
|
191 |
def clear_result():
|
192 |
"""Clears the result Image."""
|
|
|
197 |
if target_ratio == "9:16":
|
198 |
changed_width = 720
|
199 |
changed_height = 1280
|
200 |
+
return changed_width, changed_height, gr.update()
|
201 |
elif target_ratio == "16:9":
|
202 |
changed_width = 1280
|
203 |
changed_height = 720
|
204 |
+
return changed_width, changed_height, gr.update()
|
205 |
elif target_ratio == "1:1":
|
206 |
changed_width = 1024
|
207 |
changed_height = 1024
|
208 |
+
return changed_width, changed_height, gr.update()
|
209 |
elif target_ratio == "Custom":
|
|
|
210 |
return ui_width, ui_height, gr.update(open=True)
|
|
|
|
|
|
|
211 |
|
212 |
def select_the_right_preset(user_width, user_height):
|
213 |
if user_width == 720 and user_height == 1280:
|
|
|
224 |
|
225 |
def update_history(new_image, history):
|
226 |
"""Updates the history gallery with the new image."""
|
|
|
|
|
227 |
if history is None:
|
228 |
history = []
|
|
|
229 |
history.insert(0, new_image)
|
|
|
|
|
|
|
|
|
230 |
return history
|
231 |
|
232 |
+
# CSS and Title
|
233 |
css = """
|
234 |
h1 {
|
235 |
+
text-align: center;
|
236 |
+
display: block;
|
|
|
|
|
|
|
|
|
237 |
}
|
238 |
"""
|
239 |
|
240 |
title = """<h1 align="center">Diffusers Image Outpaint Lightning</h1>
|
|
|
241 |
"""
|
242 |
|
|
|
243 |
with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
|
244 |
with gr.Column():
|
245 |
gr.HTML(title)
|
246 |
|
247 |
with gr.Row():
|
248 |
+
with gr.Column():
|
249 |
input_image = gr.Image(
|
250 |
type="pil",
|
251 |
label="Input Image"
|
252 |
)
|
253 |
+
model_selection = gr.Dropdown(
|
254 |
+
choices=["RealVisXL_V5.0_Lightning", "RealVisXL_V4.0_Lightning"],
|
255 |
+
value="RealVisXL_V5.0_Lightning",
|
256 |
+
label="Select Model"
|
|
|
|
|
257 |
)
|
|
|
258 |
with gr.Row():
|
259 |
with gr.Column(scale=2):
|
260 |
+
prompt_input = gr.Textbox(label="Prompt (Optional)")
|
261 |
+
with gr.Column(scale=1):
|
262 |
+
run_button = gr.Button("Generate")
|
263 |
|
264 |
with gr.Row():
|
265 |
target_ratio = gr.Radio(
|
266 |
+
label="Expected Ratio",
|
267 |
choices=["9:16", "16:9", "1:1", "Custom"],
|
268 |
+
value="9:16",
|
269 |
scale=2
|
270 |
)
|
|
|
271 |
alignment_dropdown = gr.Dropdown(
|
272 |
choices=["Middle", "Left", "Right", "Top", "Bottom"],
|
273 |
value="Middle",
|
274 |
+
label="Alignment"
|
275 |
)
|
276 |
|
277 |
with gr.Accordion(label="Advanced settings", open=False) as settings_panel:
|
|
|
279 |
with gr.Row():
|
280 |
width_slider = gr.Slider(
|
281 |
label="Target Width",
|
282 |
+
minimum=720,
|
283 |
maximum=1536,
|
284 |
+
step=8,
|
285 |
+
value=720,
|
286 |
)
|
287 |
height_slider = gr.Slider(
|
288 |
label="Target Height",
|
289 |
+
minimum=720,
|
290 |
maximum=1536,
|
291 |
+
step=8,
|
292 |
+
value=1280,
|
293 |
)
|
|
|
294 |
num_inference_steps = gr.Slider(label="Steps", minimum=4, maximum=12, step=1, value=8)
|
|
|
295 |
with gr.Group():
|
296 |
overlap_percentage = gr.Slider(
|
297 |
label="Mask overlap (%)",
|
|
|
298 |
minimum=1,
|
299 |
maximum=50,
|
300 |
+
value=10,
|
301 |
step=1
|
302 |
)
|
|
|
303 |
with gr.Row():
|
304 |
+
overlap_top = gr.Checkbox(label="Overlap Top", value=True)
|
305 |
+
overlap_right = gr.Checkbox(label="Overlap Right", value=True)
|
306 |
+
with gr.Row():
|
307 |
+
overlap_left = gr.Checkbox(label="Overlap Left", value=True)
|
308 |
+
overlap_bottom = gr.Checkbox(label="Overlap Bottom", value=True)
|
309 |
with gr.Row():
|
310 |
resize_option = gr.Radio(
|
311 |
+
label="Resize input image",
|
|
|
312 |
choices=["Full", "50%", "33%", "25%", "Custom"],
|
313 |
+
value="Full"
|
314 |
)
|
315 |
custom_resize_percentage = gr.Slider(
|
316 |
label="Custom resize (%)",
|
|
|
318 |
maximum=100,
|
319 |
step=1,
|
320 |
value=50,
|
321 |
+
visible=False
|
322 |
)
|
323 |
+
status_text = gr.Textbox(label="Status", interactive=False)
|
324 |
gr.Examples(
|
325 |
examples=[
|
326 |
+
["./examples/example_1.webp", 1280, 720, "Middle"],
|
327 |
+
["./examples/example_2.jpg", 1440, 810, "Left"],
|
328 |
+
["./examples/example_3.jpg", 1024, 1024, "Top"],
|
329 |
+
["./examples/example_3.jpg", 1024, 1024, "Bottom"],
|
330 |
],
|
331 |
+
inputs=[input_image, width_slider, height_slider, alignment_dropdown],
|
|
|
332 |
)
|
333 |
|
334 |
+
with gr.Column():
|
335 |
result = gr.Image(
|
336 |
interactive=False,
|
337 |
label="Generated Image",
|
338 |
format="png",
|
339 |
)
|
340 |
+
history_gallery = gr.Gallery(label="History", columns=6, object_fit="contain", interactive=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
341 |
|
342 |
+
# Event handlers
|
343 |
+
model_selection.change(fn=load_model, inputs=model_selection, outputs=status_text)
|
344 |
target_ratio.change(
|
345 |
fn=preload_presets,
|
346 |
inputs=[target_ratio, width_slider, height_slider],
|
347 |
outputs=[width_slider, height_slider, settings_panel],
|
348 |
queue=False
|
349 |
)
|
|
|
|
|
350 |
width_slider.change(
|
351 |
fn=select_the_right_preset,
|
352 |
inputs=[width_slider, height_slider],
|
|
|
359 |
outputs=[target_ratio],
|
360 |
queue=False
|
361 |
)
|
|
|
|
|
362 |
resize_option.change(
|
363 |
fn=toggle_custom_resize_slider,
|
364 |
inputs=[resize_option],
|
365 |
outputs=[custom_resize_percentage],
|
366 |
queue=False
|
367 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
368 |
run_button.click(
|
369 |
fn=clear_result,
|
370 |
inputs=None,
|
371 |
+
outputs=result,
|
|
|
372 |
).then(
|
373 |
fn=infer,
|
374 |
+
inputs=[input_image, width_slider, height_slider, overlap_percentage, num_inference_steps,
|
375 |
+
resize_option, custom_resize_percentage, prompt_input, alignment_dropdown,
|
376 |
+
overlap_left, overlap_right, overlap_top, overlap_bottom],
|
377 |
+
outputs=result,
|
378 |
).then(
|
379 |
+
fn=lambda x, history: update_history(x, history),
|
380 |
+
inputs=[result, history_gallery],
|
381 |
+
outputs=history_gallery,
|
382 |
)
|
|
|
|
|
383 |
prompt_input.submit(
|
384 |
+
fn=clear_result,
|
385 |
inputs=None,
|
386 |
+
outputs=result,
|
|
|
387 |
).then(
|
388 |
fn=infer,
|
389 |
+
inputs=[input_image, width_slider, height_slider, overlap_percentage, num_inference_steps,
|
390 |
+
resize_option, custom_resize_percentage, prompt_input, alignment_dropdown,
|
391 |
+
overlap_left, overlap_right, overlap_top, overlap_bottom],
|
392 |
+
outputs=result,
|
393 |
).then(
|
394 |
+
fn=lambda x, history: update_history(x, history),
|
395 |
inputs=[result, history_gallery],
|
396 |
+
outputs=history_gallery,
|
397 |
)
|
398 |
+
demo.load(fn=load_model, inputs=model_selection, outputs=status_text)
|
399 |
|
|
|
|
|
|
|
400 |
demo.queue(max_size=20).launch(share=False, ssr_mode=False, show_error=True)
|