Spaces:
Running
on
Zero
Running
on
Zero
Upload 3 files
Browse files- Schoolboy_with_backpack.webp +0 -0
- Schoolboy_without_backpack.webp +0 -0
- app.py +490 -490
Schoolboy_with_backpack.webp
ADDED
|
Schoolboy_without_backpack.webp
ADDED
|
app.py
CHANGED
|
@@ -1,491 +1,491 @@
|
|
| 1 |
-
import os
|
| 2 |
-
# PyTorch 2.8 (temporary hack)
|
| 3 |
-
os.system('pip install --upgrade --pre --extra-index-url https://download.pytorch.org/whl/nightly/cu126 "torch<2.9" spaces')
|
| 4 |
-
|
| 5 |
-
# --- 1. Model Download and Setup (Diffusers Backend) ---
|
| 6 |
-
try:
|
| 7 |
-
import spaces
|
| 8 |
-
except:
|
| 9 |
-
class spaces():
|
| 10 |
-
def GPU(*args, **kwargs):
|
| 11 |
-
def decorator(function):
|
| 12 |
-
return lambda *dummy_args, **dummy_kwargs: function(*dummy_args, **dummy_kwargs)
|
| 13 |
-
return decorator
|
| 14 |
-
|
| 15 |
-
import torch
|
| 16 |
-
from diffusers import FlowMatchEulerDiscreteScheduler
|
| 17 |
-
from diffusers.pipelines.wan.pipeline_wan_i2v import WanImageToVideoPipeline
|
| 18 |
-
from diffusers.models.transformers.transformer_wan import WanTransformer3DModel
|
| 19 |
-
from diffusers.utils.export_utils import export_to_video
|
| 20 |
-
import gradio as gr
|
| 21 |
-
import tempfile
|
| 22 |
-
import time
|
| 23 |
-
from datetime import datetime
|
| 24 |
-
import numpy as np
|
| 25 |
-
from PIL import Image
|
| 26 |
-
import random
|
| 27 |
-
import math
|
| 28 |
-
import gc
|
| 29 |
-
from gradio_client import Client, handle_file # Import for API call
|
| 30 |
-
|
| 31 |
-
# Import the optimization function from the separate file
|
| 32 |
-
from optimization import optimize_pipeline_
|
| 33 |
-
|
| 34 |
-
# --- Constants and Model Loading ---
|
| 35 |
-
MODEL_ID = "Wan-AI/Wan2.2-I2V-A14B-Diffusers"
|
| 36 |
-
|
| 37 |
-
# --- NEW: Flexible Dimension Constants ---
|
| 38 |
-
MAX_DIMENSION = 832
|
| 39 |
-
MIN_DIMENSION = 480
|
| 40 |
-
DIMENSION_MULTIPLE = 16
|
| 41 |
-
SQUARE_SIZE = 480
|
| 42 |
-
|
| 43 |
-
MAX_SEED = np.iinfo(np.int32).max
|
| 44 |
-
|
| 45 |
-
FIXED_FPS = 24
|
| 46 |
-
MIN_FRAMES_MODEL = 8
|
| 47 |
-
MAX_FRAMES_MODEL = 81
|
| 48 |
-
|
| 49 |
-
MIN_DURATION = round(MIN_FRAMES_MODEL/FIXED_FPS, 1)
|
| 50 |
-
MAX_DURATION = round(MAX_FRAMES_MODEL/FIXED_FPS, 1)
|
| 51 |
-
|
| 52 |
-
input_image_debug_value = [None]
|
| 53 |
-
end_image_debug_value = [None]
|
| 54 |
-
prompt_debug_value = [None]
|
| 55 |
-
total_second_length_debug_value = [None]
|
| 56 |
-
|
| 57 |
-
default_negative_prompt = "Vibrant colors, overexposure, static, blurred details, subtitles, error, style, artwork, painting, image, still, overall gray, worst quality, low quality, JPEG compression residue, ugly, mutilated, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, malformed limbs, fused fingers, still image, cluttered background, three legs, many people in the background, walking backwards, overexposure, jumpcut, crossfader, "
|
| 58 |
-
|
| 59 |
-
print("Loading transformer...")
|
| 60 |
-
|
| 61 |
-
transformer = WanTransformer3DModel.from_pretrained('cbensimon/Wan2.2-I2V-A14B-bf16-Diffusers',
|
| 62 |
-
subfolder='transformer',
|
| 63 |
-
torch_dtype=torch.bfloat16,
|
| 64 |
-
device_map='cuda',
|
| 65 |
-
)
|
| 66 |
-
|
| 67 |
-
print("Loadingtransformer 2...")
|
| 68 |
-
|
| 69 |
-
transformer_2 = WanTransformer3DModel.from_pretrained('cbensimon/Wan2.2-I2V-A14B-bf16-Diffusers',
|
| 70 |
-
subfolder='transformer_2',
|
| 71 |
-
torch_dtype=torch.bfloat16,
|
| 72 |
-
device_map='cuda',
|
| 73 |
-
)
|
| 74 |
-
|
| 75 |
-
print("Loading models into memory. This may take a few minutes...")
|
| 76 |
-
|
| 77 |
-
pipe = WanImageToVideoPipeline.from_pretrained(
|
| 78 |
-
MODEL_ID,
|
| 79 |
-
transformer = transformer,
|
| 80 |
-
transformer_2 = transformer_2,
|
| 81 |
-
torch_dtype=torch.bfloat16,
|
| 82 |
-
)
|
| 83 |
-
print("Loading scheduler...")
|
| 84 |
-
pipe.scheduler = FlowMatchEulerDiscreteScheduler.from_config(pipe.scheduler.config, shift=8.0)
|
| 85 |
-
pipe.to('cuda')
|
| 86 |
-
|
| 87 |
-
print("Clean cache...")
|
| 88 |
-
for i in range(3):
|
| 89 |
-
gc.collect()
|
| 90 |
-
torch.cuda.synchronize()
|
| 91 |
-
torch.cuda.empty_cache()
|
| 92 |
-
|
| 93 |
-
print("Optimizing pipeline...")
|
| 94 |
-
|
| 95 |
-
optimize_pipeline_(pipe,
|
| 96 |
-
image=Image.new('RGB', (MAX_DIMENSION, MIN_DIMENSION)),
|
| 97 |
-
prompt='prompt',
|
| 98 |
-
height=MIN_DIMENSION,
|
| 99 |
-
width=MAX_DIMENSION,
|
| 100 |
-
num_frames=MAX_FRAMES_MODEL,
|
| 101 |
-
)
|
| 102 |
-
print("All models loaded and optimized. Gradio app is ready.")
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
# --- 2. Image Processing and Application Logic ---
|
| 106 |
-
def generate_end_frame(start_img, gen_prompt, progress=gr.Progress(track_tqdm=True)):
|
| 107 |
-
"""Calls an external Gradio API to generate an image."""
|
| 108 |
-
if start_img is None:
|
| 109 |
-
raise gr.Error("Please provide a Start Frame first.")
|
| 110 |
-
|
| 111 |
-
hf_token = os.getenv("HF_TOKEN")
|
| 112 |
-
if not hf_token:
|
| 113 |
-
raise gr.Error("HF_TOKEN not found in environment variables. Please set it in your Space secrets.")
|
| 114 |
-
|
| 115 |
-
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmpfile:
|
| 116 |
-
start_img.save(tmpfile.name)
|
| 117 |
-
tmp_path = tmpfile.name
|
| 118 |
-
|
| 119 |
-
progress(0.1, desc="Connecting to image generation API...")
|
| 120 |
-
client = Client("multimodalart/nano-banana-private")
|
| 121 |
-
|
| 122 |
-
progress(0.5, desc=f"Generating with prompt: '{gen_prompt}'...")
|
| 123 |
-
try:
|
| 124 |
-
result = client.predict(
|
| 125 |
-
prompt=gen_prompt,
|
| 126 |
-
images=[
|
| 127 |
-
{"image": handle_file(tmp_path)}
|
| 128 |
-
],
|
| 129 |
-
manual_token=hf_token,
|
| 130 |
-
api_name="/unified_image_generator"
|
| 131 |
-
)
|
| 132 |
-
finally:
|
| 133 |
-
os.remove(tmp_path)
|
| 134 |
-
|
| 135 |
-
progress(1.0, desc="Done!")
|
| 136 |
-
print(result)
|
| 137 |
-
return result
|
| 138 |
-
|
| 139 |
-
def switch_to_upload_tab():
|
| 140 |
-
"""Returns a gr.Tabs update to switch to the first tab."""
|
| 141 |
-
return gr.Tabs(selected="upload_tab")
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
def process_image_for_video(image: Image.Image) -> Image.Image:
|
| 145 |
-
"""
|
| 146 |
-
Resizes an image based on the following rules for video generation:
|
| 147 |
-
1. The longest side will be scaled down to MAX_DIMENSION if it's larger.
|
| 148 |
-
2. The shortest side will be scaled up to MIN_DIMENSION if it's smaller.
|
| 149 |
-
3. The final dimensions will be rounded to the nearest multiple of DIMENSION_MULTIPLE.
|
| 150 |
-
4. Square images are resized to a fixed SQUARE_SIZE.
|
| 151 |
-
The aspect ratio is preserved as closely as possible.
|
| 152 |
-
"""
|
| 153 |
-
width, height = image.size
|
| 154 |
-
|
| 155 |
-
# Rule 4: Handle square images
|
| 156 |
-
if width == height:
|
| 157 |
-
return image.resize((SQUARE_SIZE, SQUARE_SIZE), Image.Resampling.LANCZOS)
|
| 158 |
-
|
| 159 |
-
# Determine target dimensions while preserving aspect ratio
|
| 160 |
-
aspect_ratio = width / height
|
| 161 |
-
new_width, new_height = width, height
|
| 162 |
-
|
| 163 |
-
# Rule 1: Scale down if too large
|
| 164 |
-
if new_width > MAX_DIMENSION or new_height > MAX_DIMENSION:
|
| 165 |
-
if aspect_ratio > 1: # Landscape
|
| 166 |
-
scale = MAX_DIMENSION / new_width
|
| 167 |
-
else: # Portrait
|
| 168 |
-
scale = MAX_DIMENSION / new_height
|
| 169 |
-
new_width *= scale
|
| 170 |
-
new_height *= scale
|
| 171 |
-
|
| 172 |
-
# Rule 2: Scale up if too small
|
| 173 |
-
if new_width < MIN_DIMENSION or new_height < MIN_DIMENSION:
|
| 174 |
-
if aspect_ratio > 1: # Landscape
|
| 175 |
-
scale = MIN_DIMENSION / new_height
|
| 176 |
-
else: # Portrait
|
| 177 |
-
scale = MIN_DIMENSION / new_width
|
| 178 |
-
new_width *= scale
|
| 179 |
-
new_height *= scale
|
| 180 |
-
|
| 181 |
-
# Rule 3: Round to the nearest multiple of DIMENSION_MULTIPLE
|
| 182 |
-
final_width = int(round(new_width / DIMENSION_MULTIPLE) * DIMENSION_MULTIPLE)
|
| 183 |
-
final_height = int(round(new_height / DIMENSION_MULTIPLE) * DIMENSION_MULTIPLE)
|
| 184 |
-
|
| 185 |
-
# Ensure final dimensions are at least the minimum
|
| 186 |
-
final_width = max(final_width, MIN_DIMENSION if aspect_ratio < 1 else SQUARE_SIZE)
|
| 187 |
-
final_height = max(final_height, MIN_DIMENSION if aspect_ratio > 1 else SQUARE_SIZE)
|
| 188 |
-
|
| 189 |
-
|
| 190 |
-
return image.resize((final_width, final_height), Image.Resampling.LANCZOS)
|
| 191 |
-
|
| 192 |
-
def resize_and_crop_to_match(target_image, reference_image):
|
| 193 |
-
"""Resizes and center-crops the target image to match the reference image's dimensions."""
|
| 194 |
-
ref_width, ref_height = reference_image.size
|
| 195 |
-
target_width, target_height = target_image.size
|
| 196 |
-
scale = max(ref_width / target_width, ref_height / target_height)
|
| 197 |
-
new_width, new_height = int(target_width * scale), int(target_height * scale)
|
| 198 |
-
resized = target_image.resize((new_width, new_height), Image.Resampling.LANCZOS)
|
| 199 |
-
left, top = (new_width - ref_width) // 2, (new_height - ref_height) // 2
|
| 200 |
-
return resized.crop((left, top, left + ref_width, top + ref_height))
|
| 201 |
-
|
| 202 |
-
def generate_video(
|
| 203 |
-
start_image_pil,
|
| 204 |
-
end_image_pil,
|
| 205 |
-
prompt,
|
| 206 |
-
negative_prompt=default_negative_prompt,
|
| 207 |
-
duration_seconds=2.1,
|
| 208 |
-
steps=8,
|
| 209 |
-
guidance_scale=1,
|
| 210 |
-
guidance_scale_2=1,
|
| 211 |
-
seed=42,
|
| 212 |
-
randomize_seed=True,
|
| 213 |
-
progress=gr.Progress(track_tqdm=True)
|
| 214 |
-
):
|
| 215 |
-
start = time.time()
|
| 216 |
-
allocation_time = 120
|
| 217 |
-
factor = 1
|
| 218 |
-
|
| 219 |
-
if input_image_debug_value[0] is not None or end_image_debug_value[0] is not None or prompt_debug_value[0] is not None or total_second_length_debug_value[0] is not None:
|
| 220 |
-
start_image_pil = input_image_debug_value[0]
|
| 221 |
-
end_image_pil = end_image_debug_value[0]
|
| 222 |
-
prompt = prompt_debug_value[0]
|
| 223 |
-
duration_seconds = total_second_length_debug_value[0]
|
| 224 |
-
allocation_time = min(duration_seconds * 60 * 100, 10 * 60)
|
| 225 |
-
factor = 3.1
|
| 226 |
-
|
| 227 |
-
if start_image_pil is None or end_image_pil is None:
|
| 228 |
-
raise gr.Error("Please upload both a start and an end image.")
|
| 229 |
-
|
| 230 |
-
# Step 1: Process the start image to get our target dimensions based on the new rules.
|
| 231 |
-
processed_start_image = process_image_for_video(start_image_pil)
|
| 232 |
-
|
| 233 |
-
# Step 2: Make the end image match the *exact* dimensions of the processed start image.
|
| 234 |
-
processed_end_image = resize_and_crop_to_match(end_image_pil, processed_start_image)
|
| 235 |
-
|
| 236 |
-
target_height, target_width = processed_start_image.height, processed_start_image.width
|
| 237 |
-
|
| 238 |
-
# Handle seed and frame count
|
| 239 |
-
current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
|
| 240 |
-
num_frames = np.clip(int(round(duration_seconds * FIXED_FPS)), MIN_FRAMES_MODEL, MAX_FRAMES_MODEL)
|
| 241 |
-
|
| 242 |
-
progress(0.2, desc=f"Generating {num_frames} frames at {target_width}x{target_height} (seed: {current_seed})...")
|
| 243 |
-
|
| 244 |
-
progress(0.1, desc="Preprocessing images...")
|
| 245 |
-
output_video, download_button, seed_input = generate_video_on_gpu(
|
| 246 |
-
start_image_pil,
|
| 247 |
-
end_image_pil,
|
| 248 |
-
prompt,
|
| 249 |
-
negative_prompt,
|
| 250 |
-
duration_seconds,
|
| 251 |
-
steps,
|
| 252 |
-
guidance_scale,
|
| 253 |
-
guidance_scale_2,
|
| 254 |
-
seed,
|
| 255 |
-
randomize_seed,
|
| 256 |
-
progress,
|
| 257 |
-
allocation_time,
|
| 258 |
-
factor,
|
| 259 |
-
target_height,
|
| 260 |
-
target_width,
|
| 261 |
-
current_seed,
|
| 262 |
-
num_frames,
|
| 263 |
-
processed_start_image,
|
| 264 |
-
processed_end_image
|
| 265 |
-
)
|
| 266 |
-
progress(1.0, desc="Done!")
|
| 267 |
-
end = time.time()
|
| 268 |
-
secondes = int(end - start)
|
| 269 |
-
minutes = math.floor(secondes / 60)
|
| 270 |
-
secondes = secondes - (minutes * 60)
|
| 271 |
-
hours = math.floor(minutes / 60)
|
| 272 |
-
minutes = minutes - (hours * 60)
|
| 273 |
-
information = ("Start the process again if you want a different result. " if randomize_seed else "") + \
|
| 274 |
-
"The video been generated in " + \
|
| 275 |
-
((str(hours) + " h, ") if hours != 0 else "") + \
|
| 276 |
-
((str(minutes) + " min, ") if hours != 0 or minutes != 0 else "") + \
|
| 277 |
-
str(secondes) + " sec. " + \
|
| 278 |
-
"The video resolution is " + str(target_width) + \
|
| 279 |
-
" pixels large and " + str(target_height) + \
|
| 280 |
-
" pixels high, so a resolution of " + f'{target_width * target_height:,}' + " pixels."
|
| 281 |
-
return [output_video, download_button, seed_input, gr.update(value = information, visible = True)]
|
| 282 |
-
|
| 283 |
-
def get_duration(
|
| 284 |
-
start_image_pil,
|
| 285 |
-
end_image_pil,
|
| 286 |
-
prompt,
|
| 287 |
-
negative_prompt,
|
| 288 |
-
duration_seconds,
|
| 289 |
-
steps,
|
| 290 |
-
guidance_scale,
|
| 291 |
-
guidance_scale_2,
|
| 292 |
-
seed,
|
| 293 |
-
randomize_seed,
|
| 294 |
-
progress,
|
| 295 |
-
allocation_time,
|
| 296 |
-
factor,
|
| 297 |
-
target_height,
|
| 298 |
-
target_width,
|
| 299 |
-
current_seed,
|
| 300 |
-
num_frames,
|
| 301 |
-
processed_start_image,
|
| 302 |
-
processed_end_image
|
| 303 |
-
):
|
| 304 |
-
return allocation_time
|
| 305 |
-
|
| 306 |
-
@spaces.GPU(duration=get_duration)
|
| 307 |
-
def generate_video_on_gpu(
|
| 308 |
-
start_image_pil,
|
| 309 |
-
end_image_pil,
|
| 310 |
-
prompt,
|
| 311 |
-
negative_prompt,
|
| 312 |
-
duration_seconds,
|
| 313 |
-
steps,
|
| 314 |
-
guidance_scale,
|
| 315 |
-
guidance_scale_2,
|
| 316 |
-
seed,
|
| 317 |
-
randomize_seed,
|
| 318 |
-
progress,
|
| 319 |
-
allocation_time,
|
| 320 |
-
factor,
|
| 321 |
-
target_height,
|
| 322 |
-
target_width,
|
| 323 |
-
current_seed,
|
| 324 |
-
num_frames,
|
| 325 |
-
processed_start_image,
|
| 326 |
-
processed_end_image
|
| 327 |
-
):
|
| 328 |
-
"""
|
| 329 |
-
Generates a video by interpolating between a start and end image, guided by a text prompt,
|
| 330 |
-
using the diffusers Wan2.2 pipeline.
|
| 331 |
-
"""
|
| 332 |
-
print("Generate a video with the prompt: " + prompt)
|
| 333 |
-
|
| 334 |
-
output_frames_list = pipe(
|
| 335 |
-
image=processed_start_image,
|
| 336 |
-
last_image=processed_end_image,
|
| 337 |
-
prompt=prompt,
|
| 338 |
-
negative_prompt=negative_prompt,
|
| 339 |
-
height=target_height,
|
| 340 |
-
width=target_width,
|
| 341 |
-
num_frames=int(num_frames * factor),
|
| 342 |
-
guidance_scale=float(guidance_scale),
|
| 343 |
-
guidance_scale_2=float(guidance_scale_2),
|
| 344 |
-
num_inference_steps=int(steps),
|
| 345 |
-
generator=torch.Generator(device="cuda").manual_seed(current_seed),
|
| 346 |
-
).frames[0]
|
| 347 |
-
|
| 348 |
-
progress(0.9, desc="Encoding and saving video...")
|
| 349 |
-
|
| 350 |
-
video_path = 'wan_' + datetime.now().strftime("%Y-%m-%d_%H-%M-%S.%f") + '.mp4'
|
| 351 |
-
print("Exporting video: " + video_path)
|
| 352 |
-
|
| 353 |
-
export_to_video(output_frames_list, video_path, fps=FIXED_FPS)
|
| 354 |
-
print("Video exported: " + video_path)
|
| 355 |
-
|
| 356 |
-
return video_path, gr.update(value = video_path, visible = True), current_seed
|
| 357 |
-
|
| 358 |
-
|
| 359 |
-
# --- 3. Gradio User Interface ---
|
| 360 |
-
|
| 361 |
-
with gr.Blocks() as app:
|
| 362 |
-
gr.Markdown("# Wan 2.2 First/Last Frame Video Fast")
|
| 363 |
-
gr.Markdown("Based on the [Wan 2.2 First/Last Frame workflow](https://www.reddit.com/r/StableDiffusion/comments/1me4306/psa_wan_22_does_first_frame_last_frame_out_of_the/), applied to 🧨 Diffusers + [lightx2v/Wan2.2-Lightning](https://huggingface.co/lightx2v/Wan2.2-Lightning) 8-step LoRA")
|
| 364 |
-
|
| 365 |
-
with gr.Row(elem_id="general_items"):
|
| 366 |
-
with gr.Column():
|
| 367 |
-
with gr.Group(elem_id="group_all"):
|
| 368 |
-
with gr.Row():
|
| 369 |
-
start_image = gr.Image(type="pil", label="Start Frame", sources=["upload", "clipboard"])
|
| 370 |
-
# Capture the Tabs component in a variable and assign IDs to tabs
|
| 371 |
-
with gr.Tabs(elem_id="group_tabs") as tabs:
|
| 372 |
-
with gr.TabItem("Upload", id="upload_tab"):
|
| 373 |
-
end_image = gr.Image(type="pil", label="End Frame", sources=["upload", "clipboard"])
|
| 374 |
-
with gr.TabItem("Generate", id="generate_tab"):
|
| 375 |
-
generate_5seconds = gr.Button("Generate scene 5 seconds in the future", elem_id="fivesec")
|
| 376 |
-
gr.Markdown("Generate a custom end-frame with an edit model like [Nano Banana](https://huggingface.co/spaces/multimodalart/nano-banana) or [Qwen Image Edit](https://huggingface.co/spaces/multimodalart/Qwen-Image-Edit-Fast)", elem_id="or_item")
|
| 377 |
-
prompt = gr.Textbox(label="Prompt", info="Describe the transition between the two images")
|
| 378 |
-
|
| 379 |
-
with gr.Accordion("Advanced Settings", open=False):
|
| 380 |
-
duration_seconds_input = gr.Slider(minimum=MIN_DURATION, maximum=MAX_DURATION, step=0.1, value=2.1, label="Video Duration (seconds)", info=f"Clamped to model's {MIN_FRAMES_MODEL}-{MAX_FRAMES_MODEL} frames at {FIXED_FPS}fps.")
|
| 381 |
-
negative_prompt_input = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
|
| 382 |
-
steps_slider = gr.Slider(minimum=1, maximum=30, step=1, value=8, label="Inference Steps")
|
| 383 |
-
guidance_scale_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1.0, label="Guidance Scale - high noise")
|
| 384 |
-
guidance_scale_2_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1.0, label="Guidance Scale - low noise")
|
| 385 |
-
with gr.Row():
|
| 386 |
-
seed_input = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42)
|
| 387 |
-
randomize_seed_checkbox = gr.Checkbox(label="Randomize seed", value=True)
|
| 388 |
-
|
| 389 |
-
generate_button = gr.Button("Generate Video", variant="primary")
|
| 390 |
-
|
| 391 |
-
with gr.Column():
|
| 392 |
-
output_video = gr.Video(label="Generated Video", autoplay = True, loop = True)
|
| 393 |
-
download_button = gr.DownloadButton(label="Download", visible = True)
|
| 394 |
-
video_information = gr.HTML(value = "", visible = True)
|
| 395 |
-
|
| 396 |
-
# Main video generation button
|
| 397 |
-
ui_inputs = [
|
| 398 |
-
start_image,
|
| 399 |
-
end_image,
|
| 400 |
-
prompt,
|
| 401 |
-
negative_prompt_input,
|
| 402 |
-
duration_seconds_input,
|
| 403 |
-
steps_slider,
|
| 404 |
-
guidance_scale_input,
|
| 405 |
-
guidance_scale_2_input,
|
| 406 |
-
seed_input,
|
| 407 |
-
randomize_seed_checkbox
|
| 408 |
-
]
|
| 409 |
-
ui_outputs = [output_video, download_button, seed_input, video_information]
|
| 410 |
-
|
| 411 |
-
generate_button.click(
|
| 412 |
-
fn=generate_video,
|
| 413 |
-
inputs=ui_inputs,
|
| 414 |
-
outputs=ui_outputs
|
| 415 |
-
)
|
| 416 |
-
|
| 417 |
-
generate_5seconds.click(
|
| 418 |
-
fn=switch_to_upload_tab,
|
| 419 |
-
inputs=None,
|
| 420 |
-
outputs=[tabs]
|
| 421 |
-
).then(
|
| 422 |
-
fn=lambda img: generate_end_frame(img, "this image is a still frame from a movie. generate a new frame with what happens on this scene 5 seconds in the future"),
|
| 423 |
-
inputs=[start_image],
|
| 424 |
-
outputs=[end_image]
|
| 425 |
-
).success(
|
| 426 |
-
fn=generate_video,
|
| 427 |
-
inputs=ui_inputs,
|
| 428 |
-
outputs=ui_outputs
|
| 429 |
-
)
|
| 430 |
-
|
| 431 |
-
with gr.Row(visible=False):
|
| 432 |
-
input_image_debug=gr.Image(type="pil", label="Image Debug")
|
| 433 |
-
end_image_debug=gr.Image(type="pil", label="End Image Debug")
|
| 434 |
-
prompt_debug=gr.Textbox(label="Prompt Debug")
|
| 435 |
-
total_second_length_debug=gr.Slider(label="Additional Video Length to Generate (seconds) Debug", minimum=1, maximum=120, value=10, step=0.1)
|
| 436 |
-
gr.Examples(
|
| 437 |
-
examples=[["ugly_sonic.jpeg", "squatting_sonic.png", "the character dodges the missiles"]],
|
| 438 |
-
inputs=[start_image, end_image, prompt],
|
| 439 |
-
outputs=ui_outputs,
|
| 440 |
-
fn=generate_video,
|
| 441 |
-
run_on_click=True,
|
| 442 |
-
cache_examples=True,
|
| 443 |
-
cache_mode='lazy',
|
| 444 |
-
)
|
| 445 |
-
|
| 446 |
-
gr.Examples(
|
| 447 |
-
label = "Examples from demo",
|
| 448 |
-
examples = [
|
| 449 |
-
["poli_tower.png", "tower_takes_off.png", "
|
| 450 |
-
["ugly_sonic.jpeg", "squatting_sonic.png", "
|
| 451 |
-
["
|
| 452 |
-
],
|
| 453 |
-
inputs = [start_image, end_image, prompt],
|
| 454 |
-
outputs = ui_outputs,
|
| 455 |
-
fn = generate_video,
|
| 456 |
-
cache_examples = False,
|
| 457 |
-
)
|
| 458 |
-
|
| 459 |
-
def handle_field_debug_change(input_image_debug_data, end_image_debug_data, prompt_debug_data, total_second_length_debug_data):
|
| 460 |
-
input_image_debug_value[0] = input_image_debug_data
|
| 461 |
-
end_image_debug_value[0] = end_image_debug_data
|
| 462 |
-
prompt_debug_value[0] = prompt_debug_data
|
| 463 |
-
total_second_length_debug_value[0] = total_second_length_debug_data
|
| 464 |
-
return []
|
| 465 |
-
|
| 466 |
-
input_image_debug.upload(
|
| 467 |
-
fn=handle_field_debug_change,
|
| 468 |
-
inputs=[input_image_debug, end_image_debug, prompt_debug, total_second_length_debug],
|
| 469 |
-
outputs=[]
|
| 470 |
-
)
|
| 471 |
-
|
| 472 |
-
end_image_debug.upload(
|
| 473 |
-
fn=handle_field_debug_change,
|
| 474 |
-
inputs=[input_image_debug, end_image_debug, prompt_debug, total_second_length_debug],
|
| 475 |
-
outputs=[]
|
| 476 |
-
)
|
| 477 |
-
|
| 478 |
-
prompt_debug.change(
|
| 479 |
-
fn=handle_field_debug_change,
|
| 480 |
-
inputs=[input_image_debug, end_image_debug, prompt_debug, total_second_length_debug],
|
| 481 |
-
outputs=[]
|
| 482 |
-
)
|
| 483 |
-
|
| 484 |
-
total_second_length_debug.change(
|
| 485 |
-
fn=handle_field_debug_change,
|
| 486 |
-
inputs=[input_image_debug, end_image_debug, prompt_debug, total_second_length_debug],
|
| 487 |
-
outputs=[]
|
| 488 |
-
)
|
| 489 |
-
|
| 490 |
-
if __name__ == "__main__":
|
| 491 |
app.launch(mcp_server=True, share=True)
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
# PyTorch 2.8 (temporary hack)
|
| 3 |
+
os.system('pip install --upgrade --pre --extra-index-url https://download.pytorch.org/whl/nightly/cu126 "torch<2.9" spaces')
|
| 4 |
+
|
| 5 |
+
# --- 1. Model Download and Setup (Diffusers Backend) ---
|
| 6 |
+
try:
|
| 7 |
+
import spaces
|
| 8 |
+
except:
|
| 9 |
+
class spaces():
|
| 10 |
+
def GPU(*args, **kwargs):
|
| 11 |
+
def decorator(function):
|
| 12 |
+
return lambda *dummy_args, **dummy_kwargs: function(*dummy_args, **dummy_kwargs)
|
| 13 |
+
return decorator
|
| 14 |
+
|
| 15 |
+
import torch
|
| 16 |
+
from diffusers import FlowMatchEulerDiscreteScheduler
|
| 17 |
+
from diffusers.pipelines.wan.pipeline_wan_i2v import WanImageToVideoPipeline
|
| 18 |
+
from diffusers.models.transformers.transformer_wan import WanTransformer3DModel
|
| 19 |
+
from diffusers.utils.export_utils import export_to_video
|
| 20 |
+
import gradio as gr
|
| 21 |
+
import tempfile
|
| 22 |
+
import time
|
| 23 |
+
from datetime import datetime
|
| 24 |
+
import numpy as np
|
| 25 |
+
from PIL import Image
|
| 26 |
+
import random
|
| 27 |
+
import math
|
| 28 |
+
import gc
|
| 29 |
+
from gradio_client import Client, handle_file # Import for API call
|
| 30 |
+
|
| 31 |
+
# Import the optimization function from the separate file
|
| 32 |
+
from optimization import optimize_pipeline_
|
| 33 |
+
|
| 34 |
+
# --- Constants and Model Loading ---
|
| 35 |
+
MODEL_ID = "Wan-AI/Wan2.2-I2V-A14B-Diffusers"
|
| 36 |
+
|
| 37 |
+
# --- NEW: Flexible Dimension Constants ---
|
| 38 |
+
MAX_DIMENSION = 832
|
| 39 |
+
MIN_DIMENSION = 480
|
| 40 |
+
DIMENSION_MULTIPLE = 16
|
| 41 |
+
SQUARE_SIZE = 480
|
| 42 |
+
|
| 43 |
+
MAX_SEED = np.iinfo(np.int32).max
|
| 44 |
+
|
| 45 |
+
FIXED_FPS = 24
|
| 46 |
+
MIN_FRAMES_MODEL = 8
|
| 47 |
+
MAX_FRAMES_MODEL = 81
|
| 48 |
+
|
| 49 |
+
MIN_DURATION = round(MIN_FRAMES_MODEL/FIXED_FPS, 1)
|
| 50 |
+
MAX_DURATION = round(MAX_FRAMES_MODEL/FIXED_FPS, 1)
|
| 51 |
+
|
| 52 |
+
input_image_debug_value = [None]
|
| 53 |
+
end_image_debug_value = [None]
|
| 54 |
+
prompt_debug_value = [None]
|
| 55 |
+
total_second_length_debug_value = [None]
|
| 56 |
+
|
| 57 |
+
default_negative_prompt = "Vibrant colors, overexposure, static, blurred details, subtitles, error, style, artwork, painting, image, still, overall gray, worst quality, low quality, JPEG compression residue, ugly, mutilated, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, malformed limbs, fused fingers, still image, cluttered background, three legs, many people in the background, walking backwards, overexposure, jumpcut, crossfader, "
|
| 58 |
+
|
| 59 |
+
print("Loading transformer...")
|
| 60 |
+
|
| 61 |
+
transformer = WanTransformer3DModel.from_pretrained('cbensimon/Wan2.2-I2V-A14B-bf16-Diffusers',
|
| 62 |
+
subfolder='transformer',
|
| 63 |
+
torch_dtype=torch.bfloat16,
|
| 64 |
+
device_map='cuda',
|
| 65 |
+
)
|
| 66 |
+
|
| 67 |
+
print("Loadingtransformer 2...")
|
| 68 |
+
|
| 69 |
+
transformer_2 = WanTransformer3DModel.from_pretrained('cbensimon/Wan2.2-I2V-A14B-bf16-Diffusers',
|
| 70 |
+
subfolder='transformer_2',
|
| 71 |
+
torch_dtype=torch.bfloat16,
|
| 72 |
+
device_map='cuda',
|
| 73 |
+
)
|
| 74 |
+
|
| 75 |
+
print("Loading models into memory. This may take a few minutes...")
|
| 76 |
+
|
| 77 |
+
pipe = WanImageToVideoPipeline.from_pretrained(
|
| 78 |
+
MODEL_ID,
|
| 79 |
+
transformer = transformer,
|
| 80 |
+
transformer_2 = transformer_2,
|
| 81 |
+
torch_dtype=torch.bfloat16,
|
| 82 |
+
)
|
| 83 |
+
print("Loading scheduler...")
|
| 84 |
+
pipe.scheduler = FlowMatchEulerDiscreteScheduler.from_config(pipe.scheduler.config, shift=8.0)
|
| 85 |
+
pipe.to('cuda')
|
| 86 |
+
|
| 87 |
+
print("Clean cache...")
|
| 88 |
+
for i in range(3):
|
| 89 |
+
gc.collect()
|
| 90 |
+
torch.cuda.synchronize()
|
| 91 |
+
torch.cuda.empty_cache()
|
| 92 |
+
|
| 93 |
+
print("Optimizing pipeline...")
|
| 94 |
+
|
| 95 |
+
optimize_pipeline_(pipe,
|
| 96 |
+
image=Image.new('RGB', (MAX_DIMENSION, MIN_DIMENSION)),
|
| 97 |
+
prompt='prompt',
|
| 98 |
+
height=MIN_DIMENSION,
|
| 99 |
+
width=MAX_DIMENSION,
|
| 100 |
+
num_frames=MAX_FRAMES_MODEL,
|
| 101 |
+
)
|
| 102 |
+
print("All models loaded and optimized. Gradio app is ready.")
|
| 103 |
+
|
| 104 |
+
|
| 105 |
+
# --- 2. Image Processing and Application Logic ---
|
| 106 |
+
def generate_end_frame(start_img, gen_prompt, progress=gr.Progress(track_tqdm=True)):
|
| 107 |
+
"""Calls an external Gradio API to generate an image."""
|
| 108 |
+
if start_img is None:
|
| 109 |
+
raise gr.Error("Please provide a Start Frame first.")
|
| 110 |
+
|
| 111 |
+
hf_token = os.getenv("HF_TOKEN")
|
| 112 |
+
if not hf_token:
|
| 113 |
+
raise gr.Error("HF_TOKEN not found in environment variables. Please set it in your Space secrets.")
|
| 114 |
+
|
| 115 |
+
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmpfile:
|
| 116 |
+
start_img.save(tmpfile.name)
|
| 117 |
+
tmp_path = tmpfile.name
|
| 118 |
+
|
| 119 |
+
progress(0.1, desc="Connecting to image generation API...")
|
| 120 |
+
client = Client("multimodalart/nano-banana-private")
|
| 121 |
+
|
| 122 |
+
progress(0.5, desc=f"Generating with prompt: '{gen_prompt}'...")
|
| 123 |
+
try:
|
| 124 |
+
result = client.predict(
|
| 125 |
+
prompt=gen_prompt,
|
| 126 |
+
images=[
|
| 127 |
+
{"image": handle_file(tmp_path)}
|
| 128 |
+
],
|
| 129 |
+
manual_token=hf_token,
|
| 130 |
+
api_name="/unified_image_generator"
|
| 131 |
+
)
|
| 132 |
+
finally:
|
| 133 |
+
os.remove(tmp_path)
|
| 134 |
+
|
| 135 |
+
progress(1.0, desc="Done!")
|
| 136 |
+
print(result)
|
| 137 |
+
return result
|
| 138 |
+
|
| 139 |
+
def switch_to_upload_tab():
|
| 140 |
+
"""Returns a gr.Tabs update to switch to the first tab."""
|
| 141 |
+
return gr.Tabs(selected="upload_tab")
|
| 142 |
+
|
| 143 |
+
|
| 144 |
+
def process_image_for_video(image: Image.Image) -> Image.Image:
|
| 145 |
+
"""
|
| 146 |
+
Resizes an image based on the following rules for video generation:
|
| 147 |
+
1. The longest side will be scaled down to MAX_DIMENSION if it's larger.
|
| 148 |
+
2. The shortest side will be scaled up to MIN_DIMENSION if it's smaller.
|
| 149 |
+
3. The final dimensions will be rounded to the nearest multiple of DIMENSION_MULTIPLE.
|
| 150 |
+
4. Square images are resized to a fixed SQUARE_SIZE.
|
| 151 |
+
The aspect ratio is preserved as closely as possible.
|
| 152 |
+
"""
|
| 153 |
+
width, height = image.size
|
| 154 |
+
|
| 155 |
+
# Rule 4: Handle square images
|
| 156 |
+
if width == height:
|
| 157 |
+
return image.resize((SQUARE_SIZE, SQUARE_SIZE), Image.Resampling.LANCZOS)
|
| 158 |
+
|
| 159 |
+
# Determine target dimensions while preserving aspect ratio
|
| 160 |
+
aspect_ratio = width / height
|
| 161 |
+
new_width, new_height = width, height
|
| 162 |
+
|
| 163 |
+
# Rule 1: Scale down if too large
|
| 164 |
+
if new_width > MAX_DIMENSION or new_height > MAX_DIMENSION:
|
| 165 |
+
if aspect_ratio > 1: # Landscape
|
| 166 |
+
scale = MAX_DIMENSION / new_width
|
| 167 |
+
else: # Portrait
|
| 168 |
+
scale = MAX_DIMENSION / new_height
|
| 169 |
+
new_width *= scale
|
| 170 |
+
new_height *= scale
|
| 171 |
+
|
| 172 |
+
# Rule 2: Scale up if too small
|
| 173 |
+
if new_width < MIN_DIMENSION or new_height < MIN_DIMENSION:
|
| 174 |
+
if aspect_ratio > 1: # Landscape
|
| 175 |
+
scale = MIN_DIMENSION / new_height
|
| 176 |
+
else: # Portrait
|
| 177 |
+
scale = MIN_DIMENSION / new_width
|
| 178 |
+
new_width *= scale
|
| 179 |
+
new_height *= scale
|
| 180 |
+
|
| 181 |
+
# Rule 3: Round to the nearest multiple of DIMENSION_MULTIPLE
|
| 182 |
+
final_width = int(round(new_width / DIMENSION_MULTIPLE) * DIMENSION_MULTIPLE)
|
| 183 |
+
final_height = int(round(new_height / DIMENSION_MULTIPLE) * DIMENSION_MULTIPLE)
|
| 184 |
+
|
| 185 |
+
# Ensure final dimensions are at least the minimum
|
| 186 |
+
final_width = max(final_width, MIN_DIMENSION if aspect_ratio < 1 else SQUARE_SIZE)
|
| 187 |
+
final_height = max(final_height, MIN_DIMENSION if aspect_ratio > 1 else SQUARE_SIZE)
|
| 188 |
+
|
| 189 |
+
|
| 190 |
+
return image.resize((final_width, final_height), Image.Resampling.LANCZOS)
|
| 191 |
+
|
| 192 |
+
def resize_and_crop_to_match(target_image, reference_image):
|
| 193 |
+
"""Resizes and center-crops the target image to match the reference image's dimensions."""
|
| 194 |
+
ref_width, ref_height = reference_image.size
|
| 195 |
+
target_width, target_height = target_image.size
|
| 196 |
+
scale = max(ref_width / target_width, ref_height / target_height)
|
| 197 |
+
new_width, new_height = int(target_width * scale), int(target_height * scale)
|
| 198 |
+
resized = target_image.resize((new_width, new_height), Image.Resampling.LANCZOS)
|
| 199 |
+
left, top = (new_width - ref_width) // 2, (new_height - ref_height) // 2
|
| 200 |
+
return resized.crop((left, top, left + ref_width, top + ref_height))
|
| 201 |
+
|
| 202 |
+
def generate_video(
|
| 203 |
+
start_image_pil,
|
| 204 |
+
end_image_pil,
|
| 205 |
+
prompt,
|
| 206 |
+
negative_prompt=default_negative_prompt,
|
| 207 |
+
duration_seconds=2.1,
|
| 208 |
+
steps=8,
|
| 209 |
+
guidance_scale=1,
|
| 210 |
+
guidance_scale_2=1,
|
| 211 |
+
seed=42,
|
| 212 |
+
randomize_seed=True,
|
| 213 |
+
progress=gr.Progress(track_tqdm=True)
|
| 214 |
+
):
|
| 215 |
+
start = time.time()
|
| 216 |
+
allocation_time = 120
|
| 217 |
+
factor = 1
|
| 218 |
+
|
| 219 |
+
if input_image_debug_value[0] is not None or end_image_debug_value[0] is not None or prompt_debug_value[0] is not None or total_second_length_debug_value[0] is not None:
|
| 220 |
+
start_image_pil = input_image_debug_value[0]
|
| 221 |
+
end_image_pil = end_image_debug_value[0]
|
| 222 |
+
prompt = prompt_debug_value[0]
|
| 223 |
+
duration_seconds = total_second_length_debug_value[0]
|
| 224 |
+
allocation_time = min(duration_seconds * 60 * 100, 10 * 60)
|
| 225 |
+
factor = 3.1
|
| 226 |
+
|
| 227 |
+
if start_image_pil is None or end_image_pil is None:
|
| 228 |
+
raise gr.Error("Please upload both a start and an end image.")
|
| 229 |
+
|
| 230 |
+
# Step 1: Process the start image to get our target dimensions based on the new rules.
|
| 231 |
+
processed_start_image = process_image_for_video(start_image_pil)
|
| 232 |
+
|
| 233 |
+
# Step 2: Make the end image match the *exact* dimensions of the processed start image.
|
| 234 |
+
processed_end_image = resize_and_crop_to_match(end_image_pil, processed_start_image)
|
| 235 |
+
|
| 236 |
+
target_height, target_width = processed_start_image.height, processed_start_image.width
|
| 237 |
+
|
| 238 |
+
# Handle seed and frame count
|
| 239 |
+
current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
|
| 240 |
+
num_frames = np.clip(int(round(duration_seconds * FIXED_FPS)), MIN_FRAMES_MODEL, MAX_FRAMES_MODEL)
|
| 241 |
+
|
| 242 |
+
progress(0.2, desc=f"Generating {num_frames} frames at {target_width}x{target_height} (seed: {current_seed})...")
|
| 243 |
+
|
| 244 |
+
progress(0.1, desc="Preprocessing images...")
|
| 245 |
+
output_video, download_button, seed_input = generate_video_on_gpu(
|
| 246 |
+
start_image_pil,
|
| 247 |
+
end_image_pil,
|
| 248 |
+
prompt,
|
| 249 |
+
negative_prompt,
|
| 250 |
+
duration_seconds,
|
| 251 |
+
steps,
|
| 252 |
+
guidance_scale,
|
| 253 |
+
guidance_scale_2,
|
| 254 |
+
seed,
|
| 255 |
+
randomize_seed,
|
| 256 |
+
progress,
|
| 257 |
+
allocation_time,
|
| 258 |
+
factor,
|
| 259 |
+
target_height,
|
| 260 |
+
target_width,
|
| 261 |
+
current_seed,
|
| 262 |
+
num_frames,
|
| 263 |
+
processed_start_image,
|
| 264 |
+
processed_end_image
|
| 265 |
+
)
|
| 266 |
+
progress(1.0, desc="Done!")
|
| 267 |
+
end = time.time()
|
| 268 |
+
secondes = int(end - start)
|
| 269 |
+
minutes = math.floor(secondes / 60)
|
| 270 |
+
secondes = secondes - (minutes * 60)
|
| 271 |
+
hours = math.floor(minutes / 60)
|
| 272 |
+
minutes = minutes - (hours * 60)
|
| 273 |
+
information = ("Start the process again if you want a different result. " if randomize_seed else "") + \
|
| 274 |
+
"The video been generated in " + \
|
| 275 |
+
((str(hours) + " h, ") if hours != 0 else "") + \
|
| 276 |
+
((str(minutes) + " min, ") if hours != 0 or minutes != 0 else "") + \
|
| 277 |
+
str(secondes) + " sec. " + \
|
| 278 |
+
"The video resolution is " + str(target_width) + \
|
| 279 |
+
" pixels large and " + str(target_height) + \
|
| 280 |
+
" pixels high, so a resolution of " + f'{target_width * target_height:,}' + " pixels."
|
| 281 |
+
return [output_video, download_button, seed_input, gr.update(value = information, visible = True)]
|
| 282 |
+
|
| 283 |
+
def get_duration(
|
| 284 |
+
start_image_pil,
|
| 285 |
+
end_image_pil,
|
| 286 |
+
prompt,
|
| 287 |
+
negative_prompt,
|
| 288 |
+
duration_seconds,
|
| 289 |
+
steps,
|
| 290 |
+
guidance_scale,
|
| 291 |
+
guidance_scale_2,
|
| 292 |
+
seed,
|
| 293 |
+
randomize_seed,
|
| 294 |
+
progress,
|
| 295 |
+
allocation_time,
|
| 296 |
+
factor,
|
| 297 |
+
target_height,
|
| 298 |
+
target_width,
|
| 299 |
+
current_seed,
|
| 300 |
+
num_frames,
|
| 301 |
+
processed_start_image,
|
| 302 |
+
processed_end_image
|
| 303 |
+
):
|
| 304 |
+
return allocation_time
|
| 305 |
+
|
| 306 |
+
@spaces.GPU(duration=get_duration)
|
| 307 |
+
def generate_video_on_gpu(
|
| 308 |
+
start_image_pil,
|
| 309 |
+
end_image_pil,
|
| 310 |
+
prompt,
|
| 311 |
+
negative_prompt,
|
| 312 |
+
duration_seconds,
|
| 313 |
+
steps,
|
| 314 |
+
guidance_scale,
|
| 315 |
+
guidance_scale_2,
|
| 316 |
+
seed,
|
| 317 |
+
randomize_seed,
|
| 318 |
+
progress,
|
| 319 |
+
allocation_time,
|
| 320 |
+
factor,
|
| 321 |
+
target_height,
|
| 322 |
+
target_width,
|
| 323 |
+
current_seed,
|
| 324 |
+
num_frames,
|
| 325 |
+
processed_start_image,
|
| 326 |
+
processed_end_image
|
| 327 |
+
):
|
| 328 |
+
"""
|
| 329 |
+
Generates a video by interpolating between a start and end image, guided by a text prompt,
|
| 330 |
+
using the diffusers Wan2.2 pipeline.
|
| 331 |
+
"""
|
| 332 |
+
print("Generate a video with the prompt: " + prompt)
|
| 333 |
+
|
| 334 |
+
output_frames_list = pipe(
|
| 335 |
+
image=processed_start_image,
|
| 336 |
+
last_image=processed_end_image,
|
| 337 |
+
prompt=prompt,
|
| 338 |
+
negative_prompt=negative_prompt,
|
| 339 |
+
height=target_height,
|
| 340 |
+
width=target_width,
|
| 341 |
+
num_frames=int(num_frames * factor),
|
| 342 |
+
guidance_scale=float(guidance_scale),
|
| 343 |
+
guidance_scale_2=float(guidance_scale_2),
|
| 344 |
+
num_inference_steps=int(steps),
|
| 345 |
+
generator=torch.Generator(device="cuda").manual_seed(current_seed),
|
| 346 |
+
).frames[0]
|
| 347 |
+
|
| 348 |
+
progress(0.9, desc="Encoding and saving video...")
|
| 349 |
+
|
| 350 |
+
video_path = 'wan_' + datetime.now().strftime("%Y-%m-%d_%H-%M-%S.%f") + '.mp4'
|
| 351 |
+
print("Exporting video: " + video_path)
|
| 352 |
+
|
| 353 |
+
export_to_video(output_frames_list, video_path, fps=FIXED_FPS)
|
| 354 |
+
print("Video exported: " + video_path)
|
| 355 |
+
|
| 356 |
+
return video_path, gr.update(value = video_path, visible = True), current_seed
|
| 357 |
+
|
| 358 |
+
|
| 359 |
+
# --- 3. Gradio User Interface ---
|
| 360 |
+
|
| 361 |
+
with gr.Blocks() as app:
|
| 362 |
+
gr.Markdown("# Wan 2.2 First/Last Frame Video Fast")
|
| 363 |
+
gr.Markdown("Based on the [Wan 2.2 First/Last Frame workflow](https://www.reddit.com/r/StableDiffusion/comments/1me4306/psa_wan_22_does_first_frame_last_frame_out_of_the/), applied to 🧨 Diffusers + [lightx2v/Wan2.2-Lightning](https://huggingface.co/lightx2v/Wan2.2-Lightning) 8-step LoRA")
|
| 364 |
+
|
| 365 |
+
with gr.Row(elem_id="general_items"):
|
| 366 |
+
with gr.Column():
|
| 367 |
+
with gr.Group(elem_id="group_all"):
|
| 368 |
+
with gr.Row():
|
| 369 |
+
start_image = gr.Image(type="pil", label="Start Frame", sources=["upload", "clipboard"])
|
| 370 |
+
# Capture the Tabs component in a variable and assign IDs to tabs
|
| 371 |
+
with gr.Tabs(elem_id="group_tabs") as tabs:
|
| 372 |
+
with gr.TabItem("Upload", id="upload_tab"):
|
| 373 |
+
end_image = gr.Image(type="pil", label="End Frame", sources=["upload", "clipboard"])
|
| 374 |
+
with gr.TabItem("Generate", id="generate_tab"):
|
| 375 |
+
generate_5seconds = gr.Button("Generate scene 5 seconds in the future", elem_id="fivesec")
|
| 376 |
+
gr.Markdown("Generate a custom end-frame with an edit model like [Nano Banana](https://huggingface.co/spaces/multimodalart/nano-banana) or [Qwen Image Edit](https://huggingface.co/spaces/multimodalart/Qwen-Image-Edit-Fast)", elem_id="or_item")
|
| 377 |
+
prompt = gr.Textbox(label="Prompt", info="Describe the transition between the two images")
|
| 378 |
+
|
| 379 |
+
with gr.Accordion("Advanced Settings", open=False):
|
| 380 |
+
duration_seconds_input = gr.Slider(minimum=MIN_DURATION, maximum=MAX_DURATION, step=0.1, value=2.1, label="Video Duration (seconds)", info=f"Clamped to model's {MIN_FRAMES_MODEL}-{MAX_FRAMES_MODEL} frames at {FIXED_FPS}fps.")
|
| 381 |
+
negative_prompt_input = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
|
| 382 |
+
steps_slider = gr.Slider(minimum=1, maximum=30, step=1, value=8, label="Inference Steps")
|
| 383 |
+
guidance_scale_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1.0, label="Guidance Scale - high noise")
|
| 384 |
+
guidance_scale_2_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1.0, label="Guidance Scale - low noise")
|
| 385 |
+
with gr.Row():
|
| 386 |
+
seed_input = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42)
|
| 387 |
+
randomize_seed_checkbox = gr.Checkbox(label="Randomize seed", value=True)
|
| 388 |
+
|
| 389 |
+
generate_button = gr.Button("Generate Video", variant="primary")
|
| 390 |
+
|
| 391 |
+
with gr.Column():
|
| 392 |
+
output_video = gr.Video(label="Generated Video", autoplay = True, loop = True)
|
| 393 |
+
download_button = gr.DownloadButton(label="Download", visible = True)
|
| 394 |
+
video_information = gr.HTML(value = "", visible = True)
|
| 395 |
+
|
| 396 |
+
# Main video generation button
|
| 397 |
+
ui_inputs = [
|
| 398 |
+
start_image,
|
| 399 |
+
end_image,
|
| 400 |
+
prompt,
|
| 401 |
+
negative_prompt_input,
|
| 402 |
+
duration_seconds_input,
|
| 403 |
+
steps_slider,
|
| 404 |
+
guidance_scale_input,
|
| 405 |
+
guidance_scale_2_input,
|
| 406 |
+
seed_input,
|
| 407 |
+
randomize_seed_checkbox
|
| 408 |
+
]
|
| 409 |
+
ui_outputs = [output_video, download_button, seed_input, video_information]
|
| 410 |
+
|
| 411 |
+
generate_button.click(
|
| 412 |
+
fn=generate_video,
|
| 413 |
+
inputs=ui_inputs,
|
| 414 |
+
outputs=ui_outputs
|
| 415 |
+
)
|
| 416 |
+
|
| 417 |
+
generate_5seconds.click(
|
| 418 |
+
fn=switch_to_upload_tab,
|
| 419 |
+
inputs=None,
|
| 420 |
+
outputs=[tabs]
|
| 421 |
+
).then(
|
| 422 |
+
fn=lambda img: generate_end_frame(img, "this image is a still frame from a movie. generate a new frame with what happens on this scene 5 seconds in the future"),
|
| 423 |
+
inputs=[start_image],
|
| 424 |
+
outputs=[end_image]
|
| 425 |
+
).success(
|
| 426 |
+
fn=generate_video,
|
| 427 |
+
inputs=ui_inputs,
|
| 428 |
+
outputs=ui_outputs
|
| 429 |
+
)
|
| 430 |
+
|
| 431 |
+
with gr.Row(visible=False):
|
| 432 |
+
input_image_debug=gr.Image(type="pil", label="Image Debug")
|
| 433 |
+
end_image_debug=gr.Image(type="pil", label="End Image Debug")
|
| 434 |
+
prompt_debug=gr.Textbox(label="Prompt Debug")
|
| 435 |
+
total_second_length_debug=gr.Slider(label="Additional Video Length to Generate (seconds) Debug", minimum=1, maximum=120, value=10, step=0.1)
|
| 436 |
+
gr.Examples(
|
| 437 |
+
examples=[["ugly_sonic.jpeg", "squatting_sonic.png", "the character dodges the missiles"]],
|
| 438 |
+
inputs=[start_image, end_image, prompt],
|
| 439 |
+
outputs=ui_outputs,
|
| 440 |
+
fn=generate_video,
|
| 441 |
+
run_on_click=True,
|
| 442 |
+
cache_examples=True,
|
| 443 |
+
cache_mode='lazy',
|
| 444 |
+
)
|
| 445 |
+
|
| 446 |
+
gr.Examples(
|
| 447 |
+
label = "Examples from demo",
|
| 448 |
+
examples = [
|
| 449 |
+
["poli_tower.png", "tower_takes_off.png", "The man turns around."],
|
| 450 |
+
["ugly_sonic.jpeg", "squatting_sonic.png", "पात्रं क्षेपणास्त्रं चकमाति।"],
|
| 451 |
+
["Schoolboy_without_backpack.webp", "Schoolboy_with_backpack.webp", "The schoolchild puts on their schoolbag."],
|
| 452 |
+
],
|
| 453 |
+
inputs = [start_image, end_image, prompt],
|
| 454 |
+
outputs = ui_outputs,
|
| 455 |
+
fn = generate_video,
|
| 456 |
+
cache_examples = False,
|
| 457 |
+
)
|
| 458 |
+
|
| 459 |
+
def handle_field_debug_change(input_image_debug_data, end_image_debug_data, prompt_debug_data, total_second_length_debug_data):
|
| 460 |
+
input_image_debug_value[0] = input_image_debug_data
|
| 461 |
+
end_image_debug_value[0] = end_image_debug_data
|
| 462 |
+
prompt_debug_value[0] = prompt_debug_data
|
| 463 |
+
total_second_length_debug_value[0] = total_second_length_debug_data
|
| 464 |
+
return []
|
| 465 |
+
|
| 466 |
+
input_image_debug.upload(
|
| 467 |
+
fn=handle_field_debug_change,
|
| 468 |
+
inputs=[input_image_debug, end_image_debug, prompt_debug, total_second_length_debug],
|
| 469 |
+
outputs=[]
|
| 470 |
+
)
|
| 471 |
+
|
| 472 |
+
end_image_debug.upload(
|
| 473 |
+
fn=handle_field_debug_change,
|
| 474 |
+
inputs=[input_image_debug, end_image_debug, prompt_debug, total_second_length_debug],
|
| 475 |
+
outputs=[]
|
| 476 |
+
)
|
| 477 |
+
|
| 478 |
+
prompt_debug.change(
|
| 479 |
+
fn=handle_field_debug_change,
|
| 480 |
+
inputs=[input_image_debug, end_image_debug, prompt_debug, total_second_length_debug],
|
| 481 |
+
outputs=[]
|
| 482 |
+
)
|
| 483 |
+
|
| 484 |
+
total_second_length_debug.change(
|
| 485 |
+
fn=handle_field_debug_change,
|
| 486 |
+
inputs=[input_image_debug, end_image_debug, prompt_debug, total_second_length_debug],
|
| 487 |
+
outputs=[]
|
| 488 |
+
)
|
| 489 |
+
|
| 490 |
+
if __name__ == "__main__":
|
| 491 |
app.launch(mcp_server=True, share=True)
|