Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -47,11 +47,11 @@ FIXED_FPS = 16
|
|
47 |
MIN_FRAMES_MODEL = 8
|
48 |
MAX_FRAMES_MODEL = 81
|
49 |
|
50 |
-
# Default prompts for different modes
|
51 |
MODE_PROMPTS = {
|
52 |
-
"
|
53 |
-
"
|
54 |
-
"
|
55 |
}
|
56 |
|
57 |
default_negative_prompt = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards, watermark, text, signature"
|
@@ -335,7 +335,7 @@ def get_duration(gallery_images, mode, prompt, height, width,
|
|
335 |
base_duration = 75
|
336 |
|
337 |
# Add extra time for background removal processing
|
338 |
-
if mode == "
|
339 |
base_duration += 30
|
340 |
|
341 |
return base_duration
|
@@ -351,7 +351,7 @@ def generate_video(gallery_images, mode, prompt, height, width,
|
|
351 |
|
352 |
Args:
|
353 |
gallery_images (list): List of PIL images from the gallery
|
354 |
-
mode (str): Processing mode - "
|
355 |
prompt (str): Text prompt describing the desired animation
|
356 |
height (int): Target height for the output video
|
357 |
width (int): Target width for the output video
|
@@ -361,7 +361,7 @@ def generate_video(gallery_images, mode, prompt, height, width,
|
|
361 |
steps (int): Number of inference steps
|
362 |
seed (int): Random seed for reproducible results
|
363 |
randomize_seed (bool): Whether to use a random seed
|
364 |
-
remove_bg (bool): Whether to remove background from images (
|
365 |
progress (gr.Progress): Gradio progress tracker
|
366 |
|
367 |
Returns:
|
@@ -370,13 +370,13 @@ def generate_video(gallery_images, mode, prompt, height, width,
|
|
370 |
if gallery_images is None or len(gallery_images) == 0:
|
371 |
raise gr.Error("Please upload at least one image to the gallery.")
|
372 |
else:
|
373 |
-
# Process images: remove background if requested (
|
374 |
processed_images = []
|
375 |
for img in gallery_images:
|
376 |
image = img[0] # Extract PIL image from gallery format
|
377 |
|
378 |
-
# Apply background removal only for
|
379 |
-
if mode == "
|
380 |
image = remove_background_from_image(image)
|
381 |
|
382 |
# Always remove alpha channels to ensure RGB format
|
@@ -385,10 +385,10 @@ def generate_video(gallery_images, mode, prompt, height, width,
|
|
385 |
|
386 |
gallery_images = processed_images
|
387 |
|
388 |
-
if mode == "
|
389 |
gallery_images = gallery_images[:2]
|
390 |
-
elif mode == "
|
391 |
-
raise gr.Error("
|
392 |
|
393 |
target_h = max(MOD_VALUE, (int(height) // MOD_VALUE) * MOD_VALUE)
|
394 |
target_w = max(MOD_VALUE, (int(width) // MOD_VALUE) * MOD_VALUE)
|
@@ -398,7 +398,7 @@ def generate_video(gallery_images, mode, prompt, height, width,
|
|
398 |
current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
|
399 |
|
400 |
# Process images based on the selected mode
|
401 |
-
if mode == "
|
402 |
frames, mask = prepare_video_and_mask_FLF2V(
|
403 |
first_img=gallery_images[0],
|
404 |
last_img=gallery_images[1],
|
@@ -407,10 +407,10 @@ def generate_video(gallery_images, mode, prompt, height, width,
|
|
407 |
num_frames=num_frames
|
408 |
)
|
409 |
reference_images = None
|
410 |
-
elif mode == "
|
411 |
frames, mask = prepare_video_and_mask_Ref2V(height=target_h, width=target_w, num_frames=num_frames)
|
412 |
reference_images = gallery_images
|
413 |
-
else: # mode == "
|
414 |
# Calculate dynamic frame indices based on number of images and frames
|
415 |
frame_indices = calculate_random2v_frame_indices(len(gallery_images), num_frames)
|
416 |
|
@@ -444,13 +444,13 @@ def generate_video(gallery_images, mode, prompt, height, width,
|
|
444 |
return video_path, current_seed
|
445 |
|
446 |
control_modes = """
|
447 |
-
**3 control modes
|
448 |
|
449 |
-
**
|
450 |
|
451 |
-
**
|
452 |
|
453 |
-
**
|
454 |
"""
|
455 |
|
456 |
with gr.Blocks() as demo:
|
@@ -473,15 +473,22 @@ with gr.Blocks() as demo:
|
|
473 |
allow_preview=True
|
474 |
)
|
475 |
|
476 |
-
# Radio button for mode selection
|
477 |
mode_radio = gr.Radio(
|
478 |
-
choices=["
|
479 |
-
value="
|
480 |
label="Control Mode",
|
481 |
-
info="
|
482 |
)
|
483 |
|
484 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
485 |
duration_seconds_input = gr.Slider(
|
486 |
minimum=round(MIN_FRAMES_MODEL/FIXED_FPS,1),
|
487 |
maximum=round(MAX_FRAMES_MODEL/FIXED_FPS,1),
|
@@ -494,13 +501,6 @@ with gr.Blocks() as demo:
|
|
494 |
with gr.Accordion("Advanced Settings", open=False):
|
495 |
negative_prompt_input = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
|
496 |
|
497 |
-
# Background removal checkbox (only for Ref2V mode)
|
498 |
-
remove_bg_checkbox = gr.Checkbox(
|
499 |
-
label="Remove Background (Ref2V mode only)",
|
500 |
-
value=False,
|
501 |
-
info="Automatically remove background from input images when using Ref2V mode"
|
502 |
-
)
|
503 |
-
|
504 |
seed_input = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42, interactive=True)
|
505 |
randomize_seed_checkbox = gr.Checkbox(label="Randomize seed", value=True, interactive=True)
|
506 |
with gr.Row():
|
@@ -516,7 +516,7 @@ with gr.Blocks() as demo:
|
|
516 |
|
517 |
# Function to update checkbox visibility based on mode
|
518 |
def update_bg_removal_visibility(mode):
|
519 |
-
return gr.update(visible=(mode == "
|
520 |
|
521 |
# Update prompt when mode changes
|
522 |
mode_radio.change(
|
@@ -545,12 +545,11 @@ with gr.Blocks() as demo:
|
|
545 |
guidance_scale_input, steps_slider, seed_input, randomize_seed_checkbox, remove_bg_checkbox
|
546 |
]
|
547 |
|
548 |
-
generate_button.click(fn=generate_video, inputs=ui_inputs, outputs=[video_output, seed_input])
|
549 |
generate_button.click(fn=generate_video, inputs=ui_inputs, outputs=[video_output, seed_input])
|
550 |
gr.Examples(
|
551 |
examples=[
|
552 |
-
[["reachy.png", "sunglasses.jpg", "gpu_hat.png"], "
|
553 |
-
[["flf2v_input_first_frame.png", "flf2v_input_last_frame.png"], "
|
554 |
],
|
555 |
inputs=[gallery_component, mode_radio, prompt_input, height_input, width_input], outputs=[video_output, seed_input], fn=generate_video, cache_examples="lazy"
|
556 |
)
|
|
|
47 |
MIN_FRAMES_MODEL = 8
|
48 |
MAX_FRAMES_MODEL = 81
|
49 |
|
50 |
+
# Default prompts for different modes - Updated with new mode names
|
51 |
MODE_PROMPTS = {
|
52 |
+
"reference": "the playful penguin picks up the green cat eye sunglasses and puts them on",
|
53 |
+
"first - last frame": "CG animation style, a small blue bird takes off from the ground, flapping its wings. The bird's feathers are delicate, with a unique pattern on its chest. The background shows a blue sky with white clouds under bright sunshine. The camera follows the bird upward, capturing its flight and the vastness of the sky from a close-up, low-angle perspective.",
|
54 |
+
"random transitions": "Various different characters appear and disappear in a fast transition video showcasting their unique features and personalities. The video is about showcasing different dance styles, with each character performing a distinct dance move. The background is a vibrant, colorful stage with dynamic lighting that changes with each dance style. The camera captures close-ups of the dancers' expressions and movements. Highly dynamic, fast-paced music video, with quick cuts and transitions between characters, cinematic, vibrant colors"
|
55 |
}
|
56 |
|
57 |
default_negative_prompt = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards, watermark, text, signature"
|
|
|
335 |
base_duration = 75
|
336 |
|
337 |
# Add extra time for background removal processing
|
338 |
+
if mode == "reference" and remove_bg: # Updated to use new mode name
|
339 |
base_duration += 30
|
340 |
|
341 |
return base_duration
|
|
|
351 |
|
352 |
Args:
|
353 |
gallery_images (list): List of PIL images from the gallery
|
354 |
+
mode (str): Processing mode - "reference", "first - last frame", or "random transitions"
|
355 |
prompt (str): Text prompt describing the desired animation
|
356 |
height (int): Target height for the output video
|
357 |
width (int): Target width for the output video
|
|
|
361 |
steps (int): Number of inference steps
|
362 |
seed (int): Random seed for reproducible results
|
363 |
randomize_seed (bool): Whether to use a random seed
|
364 |
+
remove_bg (bool): Whether to remove background from images (reference mode only)
|
365 |
progress (gr.Progress): Gradio progress tracker
|
366 |
|
367 |
Returns:
|
|
|
370 |
if gallery_images is None or len(gallery_images) == 0:
|
371 |
raise gr.Error("Please upload at least one image to the gallery.")
|
372 |
else:
|
373 |
+
# Process images: remove background if requested (reference mode only), then remove alpha channels
|
374 |
processed_images = []
|
375 |
for img in gallery_images:
|
376 |
image = img[0] # Extract PIL image from gallery format
|
377 |
|
378 |
+
# Apply background removal only for reference mode if checkbox is checked
|
379 |
+
if mode == "reference" and remove_bg: # Updated to use new mode name
|
380 |
image = remove_background_from_image(image)
|
381 |
|
382 |
# Always remove alpha channels to ensure RGB format
|
|
|
385 |
|
386 |
gallery_images = processed_images
|
387 |
|
388 |
+
if mode == "first - last frame" and len(gallery_images) >= 2: # Updated mode name
|
389 |
gallery_images = gallery_images[:2]
|
390 |
+
elif mode == "first - last frame" and len(gallery_images) < 2: # Updated mode name
|
391 |
+
raise gr.Error("First - Last Frame mode requires at least 2 images, but only {} were supplied.".format(len(gallery_images)))
|
392 |
|
393 |
target_h = max(MOD_VALUE, (int(height) // MOD_VALUE) * MOD_VALUE)
|
394 |
target_w = max(MOD_VALUE, (int(width) // MOD_VALUE) * MOD_VALUE)
|
|
|
398 |
current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
|
399 |
|
400 |
# Process images based on the selected mode
|
401 |
+
if mode == "first - last frame": # Updated mode name
|
402 |
frames, mask = prepare_video_and_mask_FLF2V(
|
403 |
first_img=gallery_images[0],
|
404 |
last_img=gallery_images[1],
|
|
|
407 |
num_frames=num_frames
|
408 |
)
|
409 |
reference_images = None
|
410 |
+
elif mode == "reference": # Updated mode name
|
411 |
frames, mask = prepare_video_and_mask_Ref2V(height=target_h, width=target_w, num_frames=num_frames)
|
412 |
reference_images = gallery_images
|
413 |
+
else: # mode == "random transitions" # Updated mode name
|
414 |
# Calculate dynamic frame indices based on number of images and frames
|
415 |
frame_indices = calculate_random2v_frame_indices(len(gallery_images), num_frames)
|
416 |
|
|
|
444 |
return video_path, current_seed
|
445 |
|
446 |
control_modes = """
|
447 |
+
**3 control modes available:**
|
448 |
|
449 |
+
**Reference** Generate a video incorporating elements from input reference images
|
450 |
|
451 |
+
**First - Last Frame** Generate a video using first and last frame conditioning defined by input images
|
452 |
|
453 |
+
**Random Transitions** Generate a video with intermediate transitions between multiple input images
|
454 |
"""
|
455 |
|
456 |
with gr.Blocks() as demo:
|
|
|
473 |
allow_preview=True
|
474 |
)
|
475 |
|
476 |
+
# Radio button for mode selection with updated names
|
477 |
mode_radio = gr.Radio(
|
478 |
+
choices=["reference", "first - last frame", "random transitions"],
|
479 |
+
value="reference",
|
480 |
label="Control Mode",
|
481 |
+
info="Reference: Reference to Video | First - Last Frame: First-Last Frame to Video | Random Transitions: Random to Video"
|
482 |
)
|
483 |
|
484 |
+
# Background removal checkbox moved here - right beneath control modes
|
485 |
+
remove_bg_checkbox = gr.Checkbox(
|
486 |
+
label="Remove Background (Reference mode only)",
|
487 |
+
value=False,
|
488 |
+
info="Automatically remove background from input images when using Reference mode"
|
489 |
+
)
|
490 |
+
|
491 |
+
prompt_input = gr.Textbox(label="Prompt", value=MODE_PROMPTS["reference"])
|
492 |
duration_seconds_input = gr.Slider(
|
493 |
minimum=round(MIN_FRAMES_MODEL/FIXED_FPS,1),
|
494 |
maximum=round(MAX_FRAMES_MODEL/FIXED_FPS,1),
|
|
|
501 |
with gr.Accordion("Advanced Settings", open=False):
|
502 |
negative_prompt_input = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
|
503 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
504 |
seed_input = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42, interactive=True)
|
505 |
randomize_seed_checkbox = gr.Checkbox(label="Randomize seed", value=True, interactive=True)
|
506 |
with gr.Row():
|
|
|
516 |
|
517 |
# Function to update checkbox visibility based on mode
|
518 |
def update_bg_removal_visibility(mode):
|
519 |
+
return gr.update(visible=(mode == "reference")) # Updated to use new mode name
|
520 |
|
521 |
# Update prompt when mode changes
|
522 |
mode_radio.change(
|
|
|
545 |
guidance_scale_input, steps_slider, seed_input, randomize_seed_checkbox, remove_bg_checkbox
|
546 |
]
|
547 |
|
|
|
548 |
generate_button.click(fn=generate_video, inputs=ui_inputs, outputs=[video_output, seed_input])
|
549 |
gr.Examples(
|
550 |
examples=[
|
551 |
+
[["reachy.png", "sunglasses.jpg", "gpu_hat.png"], "reference", "the cute robot is wearing the sunglasses and the hat that reads 'GPU poor', and moves around playfully", 480, 832],
|
552 |
+
[["flf2v_input_first_frame.png", "flf2v_input_last_frame.png"], "first - last frame", "CG animation style, a small blue bird takes off from the ground, flapping its wings. The bird's feathers are delicate, with a unique pattern on its chest. The background shows a blue sky with white clouds under bright sunshine. The camera follows the bird upward, capturing its flight and the vastness of the sky from a close-up, low-angle perspective.", 512, 512],
|
553 |
],
|
554 |
inputs=[gallery_component, mode_radio, prompt_input, height_input, width_input], outputs=[video_output, seed_input], fn=generate_video, cache_examples="lazy"
|
555 |
)
|