Upload 2 files

app.py

@@ -0,0 +1,92 @@
+import gradio as gr
+import torch
+import numpy as np
+import sys
+from PIL import Image
+from diffusers import DiffusionPipeline
+from diffusers.utils import export_to_video
+
+# Отладка
+print(f"Python: {sys.version}")
+print(f"Torch: {torch.__version__}")
+print(f"CUDA available: {torch.cuda.is_available()}")
+if torch.cuda.is_available():
+    print(f"GPU: {torch.cuda.get_device_name(0)}")
+
+# Загрузка модели Wan 2.2 (Image-to-Video)
+pipe = DiffusionPipeline.from_pretrained(
+    "Wan-AI/Wan2.2-T2V-A14B",
+    torch_dtype=torch.float16,
+    variant="fp16"
+)
+
+# Единственная безопасная оптимизация
+pipe.enable_vae_slicing()
+
+# Перемещение на GPU
+if torch.cuda.is_available():
+    pipe = pipe.to("cuda")
+    print("✓ Model loaded on GPU")
+
+def generate_video(prompt, image=None, negative_prompt="blurry, low quality, jittery"):
+    try:
+        print(f"Generating video for prompt: '{prompt}'")
+        
+        # Генерация (если есть изображение — img2vid, иначе text2vid)
+        if image is not None and isinstance(image, np.ndarray):
+            image_pil = Image.fromarray(image).convert("RGB")
+            # Для Wan 2.2 img2vid требуется специальный режим — используем text2vid как fallback
+            output = pipe(
+                prompt=prompt,
+                negative_prompt=negative_prompt,
+                num_frames=24,           # 1 секунда @ 24fps
+                height=480,              # 480p для стабильности на T4
+                width=720,
+                num_inference_steps=30,
+                generator=torch.Generator(device="cuda").manual_seed(42)
+            )
+        else:
+            # Text-to-Video (основной режим)
+            output = pipe(
+                prompt=prompt,
+                negative_prompt=negative_prompt,
+                num_frames=24,
+                height=480,
+                width=720,
+                num_inference_steps=30,
+                generator=torch.Generator(device="cuda").manual_seed(42)
+            )
+        
+        # Сохранение
+        output_path = "/tmp/output.mp4"
+        export_to_video(output.frames[0], output_path, fps=24)
+        
+        return output_path
+    
+    except Exception as e:
+        print(f"ERROR: {str(e)}")
+        import traceback
+        traceback.print_exc()
+        return None
+
+# Интерфейс
+demo = gr.Interface(
+    fn=generate_video,
+    inputs=[
+        gr.Textbox(label="Prompt (English)", value="a panda eating bamboo in a forest, cinematic, 4k"),
+        gr.Image(label="Optional Image (for img2vid)", type="numpy"),
+        gr.Textbox(label="Negative Prompt", value="blurry, low quality, jittery motion")
+    ],
+    outputs=gr.Video(label="Generated Video (720×480, 24 frames, 1 second)"),
+    title="🎥 Wan 2.2 Video Generator (Apache 2.0 License)",
+    description="✅ Commercial use allowed • 720p resolution • Based on Alibaba's open-source model",
+    examples=[
+        ["a robot dancing in cyberpunk city, neon lights, cinematic", None, "blurry"],
+        ["a cat wearing sunglasses riding a skateboard, cartoon style", None, "jittery motion"],
+        ["a spaceship flying through nebula, sci-fi, 4k", None, "low quality"]
+    ],
+    cache_examples=False
+)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,7 @@
+diffusers>=0.30.0
+transformers>=4.40.0
+accelerate>=0.30.0
+safetensors>=0.4.0
+imageio>=2.31.0
+imageio-ffmpeg>=0.4.9
+einops>=0.7.0