Update app.py

app.py

@@ -1,234 +1,243 @@
-import spaces
-import torch
-from diffusers.pipelines.wan.pipeline_wan_i2v import WanImageToVideoPipeline
-from diffusers.models.transformers.transformer_wan import WanTransformer3DModel
-from diffusers.utils.export_utils import export_to_video
-import gradio as gr
-import tempfile
-import numpy as np
-from PIL import Image
-import random
-import gc
-
-from torchao.quantization import quantize_
-from torchao.quantization import Float8DynamicActivationFloat8WeightConfig
-from torchao.quantization import Int8WeightOnlyConfig
-
-import aoti
-
-
-MODEL_ID = "Wan-AI/Wan2.2-I2V-A14B-Diffusers"
-
-MAX_DIM = 832
-MIN_DIM = 480
-SQUARE_DIM = 640
-MULTIPLE_OF = 16
-
-MAX_SEED = np.iinfo(np.int32).max
-
-FIXED_FPS = 16  # можно поменять на 24, если хочешь более плавное видео
-MIN_FRAMES_MODEL = 8
-
-MIN_DURATION = round(MIN_FRAMES_MODEL / FIXED_FPS, 1)
-DEFAULT_DURATION = 5.0  # значение по умолчанию
-
-pipe = WanImageToVideoPipeline.from_pretrained(
-    MODEL_ID,
-    transformer=WanTransformer3DModel.from_pretrained(
-        'cbensimon/Wan2.2-I2V-A14B-bf16-Diffusers',
-        subfolder='transformer',
-        torch_dtype=torch.bfloat16,
-        device_map='cuda',
-    ),
-    transformer_2=WanTransformer3DModel.from_pretrained(
-        'cbensimon/Wan2.2-I2V-A14B-bf16-Diffusers',
-        subfolder='transformer_2',
-        torch_dtype=torch.bfloat16,
-        device_map='cuda',
-    ),
-    torch_dtype=torch.bfloat16,
-).to('cuda')
-
-pipe.load_lora_weights(
-    "Kijai/WanVideo_comfy", 
-    weight_name="Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors", 
-    adapter_name="lightx2v"
-)
-kwargs_lora = {"load_into_transformer_2": True}
-pipe.load_lora_weights(
-    "Kijai/WanVideo_comfy", 
-    weight_name="Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors", 
-    adapter_name="lightx2v_2", 
-    **kwargs_lora
-)
-pipe.set_adapters(["lightx2v", "lightx2v_2"], adapter_weights=[1., 1.])
-pipe.fuse_lora(adapter_names=["lightx2v"], lora_scale=3., components=["transformer"])
-pipe.fuse_lora(adapter_names=["lightx2v_2"], lora_scale=1., components=["transformer_2"])
-pipe.unload_lora_weights()
-
-quantize_(pipe.text_encoder, Int8WeightOnlyConfig())
-quantize_(pipe.transformer, Float8DynamicActivationFloat8WeightConfig())
-quantize_(pipe.transformer_2, Float8DynamicActivationFloat8WeightConfig())
-
-aoti.aoti_blocks_load(pipe.transformer, 'zerogpu-aoti/Wan2', variant='fp8da')
-aoti.aoti_blocks_load(pipe.transformer_2, 'zerogpu-aoti/Wan2', variant='fp8da')
-
-
-default_prompt_i2v = "make this image come alive, cinematic motion, smooth animation"
-default_negative_prompt = (
-    "色调艳丽, 过曝, 静态, 细节模糊不清, 字幕, 风格, 作品, 画作, 画面, 静止, "
-    "整体发灰, 最差质量, 低质量, JPEG压缩残留, 丑陋的, 残缺的, 多余的手指, "
-    "画得不好的手部, 画得不好的脸部, 畸形的, 毁容的, 形态畸形的肢体, 手指融合, "
-    "静止不动的画面, 杂乱的背景, 三条腿, 背景人很多, 倒着走"
-)
-
-
-# ✅ Исправленная функция resize_image (больше не вызовет ошибок)
-def resize_image(image: Image.Image) -> Image.Image:
-    width, height = image.size
-
-    if width == height:
-        return image.resize((SQUARE_DIM, SQUARE_DIM), Image.LANCZOS)
-
-    aspect_ratio = width / height
-    MAX_ASPECT_RATIO = MAX_DIM / MIN_DIM 
-    MIN_ASPECT_RATIO = MIN_DIM / MAX_DIM 
-
-    # Устанавливаем безопасные значения по умолчанию
-    target_w, target_h = width, height
-    image_to_resize = image
-
-    if aspect_ratio > MAX_ASPECT_RATIO:
-        crop_width = int(round(height * MAX_ASPECT_RATIO))
-        left = (width - crop_width) // 2
-        image_to_resize = image.crop((left, 0, left + crop_width, height))
-        target_w = MAX_DIM
-        target_h = int(round(target_w / MAX_ASPECT_RATIO))
-    elif aspect_ratio < MIN_ASPECT_RATIO:
-        crop_height = int(round(width / MIN_ASPECT_RATIO))
-        top = (height - crop_height) // 2
-        image_to_resize = image.crop((0, top, width, top + crop_height))
-        target_h = MAX_DIM
-        target_w = int(round(target_h * MIN_ASPECT_RATIO))
-    else:
-        if width > height:
-            target_w = MAX_DIM
-            target_h = int(round(target_w / aspect_ratio))
-        else:
-            target_h = MAX_DIM
-            target_w = int(round(target_h * aspect_ratio))
-
-    final_w = round(target_w / MULTIPLE_OF) * MULTIPLE_OF
-    final_h = round(target_h / MULTIPLE_OF) * MULTIPLE_OF
-    final_w = max(MIN_DIM, min(MAX_DIM, final_w))
-    final_h = max(MIN_DIM, min(MAX_DIM, final_h))
-    return image_to_resize.resize((final_w, final_h), Image.LANCZOS)
-
-
-def get_num_frames(duration_seconds: float):
-    # ✅ убрано ограничение на MAX_FRAMES_MODEL
-    return 1 + int(round(duration_seconds * FIXED_FPS))
-
-
-def get_duration(
-    input_image,
-    prompt,
-    steps,
-    negative_prompt,
-    duration_seconds,
-    guidance_scale,
-    guidance_scale_2,
-    seed,
-    randomize_seed,
-    progress,
-):
-    BASE_FRAMES_HEIGHT_WIDTH = 81 * 832 * 624
-    BASE_STEP_DURATION = 15
-    width, height = resize_image(input_image).size
-    frames = get_num_frames(duration_seconds)
-    factor = frames * width * height / BASE_FRAMES_HEIGHT_WIDTH
-    step_duration = BASE_STEP_DURATION * factor ** 1.5
-    return 10 + int(steps) * step_duration
-
-
-@spaces.GPU(duration=get_duration)
-def generate_video(
-    input_image,
-    prompt,
-    steps=4,
-    negative_prompt=default_negative_prompt,
-    duration_seconds=DEFAULT_DURATION,
-    guidance_scale=1,
-    guidance_scale_2=1,    
-    seed=42,
-    randomize_seed=False,
-    progress=gr.Progress(track_tqdm=True),
-):
-    if input_image is None:
-        raise gr.Error("Please upload an input image.")
-    
-    num_frames = get_num_frames(duration_seconds)
-    current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
-    resized_image = resize_image(input_image)
-
-    output_frames_list = pipe(
-        image=resized_image,
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        height=resized_image.height,
-        width=resized_image.width,
-        num_frames=num_frames,
-        guidance_scale=float(guidance_scale),
-        guidance_scale_2=float(guidance_scale_2),
-        num_inference_steps=int(steps),
-        generator=torch.Generator(device="cuda").manual_seed(current_seed),
-    ).frames[0]
-
-    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmpfile:
-        video_path = tmpfile.name
-
-    export_to_video(output_frames_list, video_path, fps=FIXED_FPS)
-    return video_path, current_seed
-
-
-# --- Gradio Interface ---
-with gr.Blocks() as demo:
-    gr.Markdown("# 🚀 Wan 2.2 I2V (14B) — Unlimited Duration Edition 🕒")
-    gr.Markdown("Generate cinematic I2V animations without duration limits. Optimized for RTX 4090.")
-    
-    with gr.Row():
-        with gr.Column():
-            input_image_component = gr.Image(type="pil", label="Input Image")
-            prompt_input = gr.Textbox(label="Prompt", value=default_prompt_i2v)
-
-            duration_seconds_input = gr.Slider(
-                minimum=0.5,
-                maximum=60.0,  # можно поднять до 120.0
-                step=0.5,
-                value=DEFAULT_DURATION,
-                label="Duration (seconds)",
-                info=f"Each second = {FIXED_FPS} frames. Longer videos require more VRAM/time."
-            )
-            
-            with gr.Accordion("Advanced Settings", open=False):
-                negative_prompt_input = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
-                seed_input = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42, interactive=True)
-                randomize_seed_checkbox = gr.Checkbox(label="Randomize seed", value=True, interactive=True)
-                steps_slider = gr.Slider(minimum=1, maximum=30, step=1, value=6, label="Inference Steps") 
-                guidance_scale_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1, label="Guidance Scale - high noise stage")
-                guidance_scale_2_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1, label="Guidance Scale 2 - low noise stage")
-
-            generate_button = gr.Button("Generate Video", variant="primary")
-        with gr.Column():
-            video_output = gr.Video(label="Generated Video", autoplay=True, interactive=False)
-    
-    ui_inputs = [
-        input_image_component, prompt_input, steps_slider,
-        negative_prompt_input, duration_seconds_input,
-        guidance_scale_input, guidance_scale_2_input,
-        seed_input, randomize_seed_checkbox
-    ]
-    generate_button.click(fn=generate_video, inputs=ui_inputs, outputs=[video_output, seed_input])
-
-if __name__ == "__main__":
-    demo.queue().launch(mcp_server=True)
+import spaces
+import torch
+from diffusers.pipelines.wan.pipeline_wan_i2v import WanImageToVideoPipeline
+from diffusers.models.transformers.transformer_wan import WanTransformer3DModel
+from diffusers.utils.export_utils import export_to_video
+import gradio as gr
+import tempfile
+import numpy as np
+from PIL import Image
+import random
+import gc
+
+from torchao.quantization import quantize_
+from torchao.quantization import Float8DynamicActivationFloat8WeightConfig
+from torchao.quantization import Int8WeightOnlyConfig
+
+import aoti
+
+
+MODEL_ID = "Wan-AI/Wan2.2-I2V-A14B-Diffusers"
+
+MAX_DIM = 832
+MIN_DIM = 480
+SQUARE_DIM = 640
+MULTIPLE_OF = 16
+
+MAX_SEED = np.iinfo(np.int32).max
+
+FIXED_FPS = 16
+MIN_FRAMES_MODEL = 8
+
+MIN_DURATION = round(MIN_FRAMES_MODEL / FIXED_FPS, 1)
+DEFAULT_DURATION = 5.0
+
+pipe = WanImageToVideoPipeline.from_pretrained(
+    MODEL_ID,
+    transformer=WanTransformer3DModel.from_pretrained(
+        'cbensimon/Wan2.2-I2V-A14B-bf16-Diffusers',
+        subfolder='transformer',
+        torch_dtype=torch.bfloat16,
+        # 🚨 ИСПРАВЛЕНИЕ 1: Используем 'auto' для распределения по всем GPU
+        device_map='auto', 
+    ),
+    transformer_2=WanTransformer3DModel.from_pretrained(
+        'cbensimon/Wan2.2-I2V-A14B-bf16-Diffusers',
+        subfolder='transformer_2',
+        torch_dtype=torch.bfloat16,
+        # 🚨 ИСПРАВЛЕНИЕ 2: Используем 'auto' для распределения по всем GPU
+        device_map='auto',
+    ),
+    torch_dtype=torch.bfloat16,
+# 🚨 ИСПРАВЛЕНИЕ 3: Удаляем .to('cuda'), так как это отменяет распределение
+)
+
+pipe.load_lora_weights(
+    "Kijai/WanVideo_comfy", 
+    weight_name="Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors", 
+    adapter_name="lightx2v"
+)
+kwargs_lora = {"load_into_transformer_2": True}
+pipe.load_lora_weights(
+    "Kijai/WanVideo_comfy", 
+    weight_name="Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors", 
+    adapter_name="lightx2v_2", 
+    **kwargs_lora
+)
+pipe.set_adapters(["lightx2v", "lightx2v_2"], adapter_weights=[1., 1.])
+pipe.fuse_lora(adapter_names=["lightx2v"], lora_scale=3., components=["transformer"])
+pipe.fuse_lora(adapter_names=["lightx2v_2"], lora_scale=1., components=["transformer_2"])
+pipe.unload_lora_weights()
+
+# ВАЖНО: Квантизация и AOTI должны выполняться после загрузки весов,
+# но до того, как модель будет перемещена куда-либо, что и происходит.
+quantize_(pipe.text_encoder, Int8WeightOnlyConfig())
+quantize_(pipe.transformer, Float8DynamicActivationFloat8WeightConfig())
+quantize_(pipe.transformer_2, Float8DynamicActivationFloat8WeightConfig())
+
+# Перемещаем на CUDA после квантизации и распределения (device_map='auto' позаботится о распределении)
+pipe.cuda() 
+
+aoti.aoti_blocks_load(pipe.transformer, 'zerogpu-aoti/Wan2', variant='fp8da')
+aoti.aoti_blocks_load(pipe.transformer_2, 'zerogpu-aoti/Wan2', variant='fp8da')
+
+
+default_prompt_i2v = "make this image come alive, cinematic motion, smooth animation"
+default_negative_prompt = (
+    "色调艳丽, 过曝, 静态, 细节模糊不清, 字幕, 风格, 作品, 画作, 画面, 静止, "
+    "整体发灰, 最差质量, 低质量, JPEG压缩残留, 丑陋的, 残缺的, 多余的手指, "
+    "画得不好的手部, 画得不好的脸部, 畸形的, 毁容的, 形态畸形的肢体, 手指融合, "
+    "静止不动的画面, 杂乱的背景, 三条腿, 背景人很多, 倒着走"
+)
+
+
+# ✅ Исправленная функция resize_image (больше не вызовет ошибок)
+def resize_image(image: Image.Image) -> Image.Image:
+    width, height = image.size
+
+    if width == height:
+        return image.resize((SQUARE_DIM, SQUARE_DIM), Image.LANCZOS)
+
+    aspect_ratio = width / height
+    MAX_ASPECT_RATIO = MAX_DIM / MIN_DIM 
+    MIN_ASPECT_RATIO = MIN_DIM / MAX_DIM 
+
+    # Устанавливаем безопасные значения по умолчанию
+    target_w, target_h = width, height
+    image_to_resize = image
+
+    if aspect_ratio > MAX_ASPECT_RATIO:
+        crop_width = int(round(height * MAX_ASPECT_RATIO))
+        left = (width - crop_width) // 2
+        image_to_resize = image.crop((left, 0, left + crop_width, height))
+        target_w = MAX_DIM
+        target_h = int(round(target_w / MAX_ASPECT_RATIO))
+    elif aspect_ratio < MIN_ASPECT_RATIO:
+        crop_height = int(round(width / MIN_ASPECT_RATIO))
+        top = (height - crop_height) // 2
+        image_to_resize = image.crop((0, top, width, top + crop_height))
+        target_h = MAX_DIM
+        target_w = int(round(target_h * MIN_ASPECT_RATIO))
+    else:
+        if width > height:
+            target_w = MAX_DIM
+            target_h = int(round(target_w / aspect_ratio))
+        else:
+            target_h = MAX_DIM
+            target_w = int(round(target_h * aspect_ratio))
+
+    final_w = round(target_w / MULTIPLE_OF) * MULTIPLE_OF
+    final_h = round(target_h / MULTIPLE_OF) * MULTIPLE_OF
+    final_w = max(MIN_DIM, min(MAX_DIM, final_w))
+    final_h = max(MIN_DIM, min(MAX_DIM, final_h))
+    return image_to_resize.resize((final_w, final_h), Image.LANCZOS)
+
+
+def get_num_frames(duration_seconds: float):
+    return 1 + int(round(duration_seconds * FIXED_FPS))
+
+
+def get_duration(
+    input_image,
+    prompt,
+    steps,
+    negative_prompt,
+    duration_seconds,
+    guidance_scale,
+    guidance_scale_2,
+    seed,
+    randomize_seed,
+    progress,
+):
+    BASE_FRAMES_HEIGHT_WIDTH = 81 * 832 * 624
+    BASE_STEP_DURATION = 15
+    width, height = resize_image(input_image).size
+    frames = get_num_frames(duration_seconds)
+    factor = frames * width * height / BASE_FRAMES_HEIGHT_WIDTH
+    step_duration = BASE_STEP_DURATION * factor ** 1.5
+    return 10 + int(steps) * step_duration
+
+
+@spaces.GPU(duration=get_duration)
+def generate_video(
+    input_image,
+    prompt,
+    steps=4,
+    negative_prompt=default_negative_prompt,
+    duration_seconds=DEFAULT_DURATION,
+    guidance_scale=1,
+    guidance_scale_2=1,    
+    seed=42,
+    randomize_seed=False,
+    progress=gr.Progress(track_tqdm=True),
+):
+    if input_image is None:
+        raise gr.Error("Please upload an input image.")
+    
+    num_frames = get_num_frames(duration_seconds)
+    current_seed = random.randint(0, MAX_SEED) if randomize_seed else int(seed)
+    resized_image = resize_image(input_image)
+
+    # ВАЖНО: pipe уже распределен по GPU, поэтому явное указание 'cuda' не требуется, 
+    # но Generator все равно должен быть на 'cuda'.
+    output_frames_list = pipe(
+        image=resized_image,
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        height=resized_image.height,
+        width=resized_image.width,
+        num_frames=num_frames,
+        guidance_scale=float(guidance_scale),
+        guidance_scale_2=float(guidance_scale_2),
+        num_inference_steps=int(steps),
+        generator=torch.Generator(device="cuda").manual_seed(current_seed),
+    ).frames[0]
+
+    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmpfile:
+        video_path = tmpfile.name
+
+    export_to_video(output_frames_list, video_path, fps=FIXED_FPS)
+    return video_path, current_seed
+
+
+# --- Gradio Interface ---
+with gr.Blocks() as demo:
+    gr.Markdown("# 🚀 Wan 2.2 I2V (14B) — Unlimited Duration Edition 🕒")
+    gr.Markdown("Generate cinematic I2V animations without duration limits. Optimized for 4x NVIDIA L40S.")
+    
+    with gr.Row():
+        with gr.Column():
+            input_image_component = gr.Image(type="pil", label="Input Image")
+            prompt_input = gr.Textbox(label="Prompt", value=default_prompt_i2v)
+
+            duration_seconds_input = gr.Slider(
+                minimum=0.5,
+                maximum=60.0,
+                step=0.5,
+                value=DEFAULT_DURATION,
+                label="Duration (seconds)",
+                info=f"Each second = {FIXED_FPS} frames. Longer videos require more VRAM/time."
+            )
+            
+            with gr.Accordion("Advanced Settings", open=False):
+                negative_prompt_input = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
+                seed_input = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42, interactive=True)
+                randomize_seed_checkbox = gr.Checkbox(label="Randomize seed", value=True, interactive=True)
+                steps_slider = gr.Slider(minimum=1, maximum=30, step=1, value=6, label="Inference Steps") 
+                guidance_scale_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1, label="Guidance Scale - high noise stage")
+                guidance_scale_2_input = gr.Slider(minimum=0.0, maximum=10.0, step=0.5, value=1, label="Guidance Scale 2 - low noise stage")
+
+            generate_button = gr.Button("Generate Video", variant="primary")
+        with gr.Column():
+            video_output = gr.Video(label="Generated Video", autoplay=True, interactive=False)
+    
+    ui_inputs = [
+        input_image_component, prompt_input, steps_slider,
+        negative_prompt_input, duration_seconds_input,
+        guidance_scale_input, guidance_scale_2_input,
+        seed_input, randomize_seed_checkbox
+    ]
+    generate_button.click(fn=generate_video, inputs=ui_inputs, outputs=[video_output, seed_input])
+
+if __name__ == "__main__":
+    demo.queue().launch(mcp_server=True)