Update app.py

app.py

@@ -5,6 +5,15 @@ import gradio as gr
 from huggingface_hub import snapshot_download
 from huggingface_hub import snapshot_download
 import spaces
+# Add PusaV1 to path to resolve diffsynth imports
+sys.path.append(os.path.abspath("PusaV1"))
+
+# Import the actual model runner
+from diffsynth import ModelManager, WanVideoPusaPipeline, save_video
+
+# Define paths
+WAN_MODEL_DIR = "./model_zoo/Wan2.1-T2V-14B"
+LORA_PATH = "./model_zoo/PusaV1/pusa_v1.pt"
 
 MODEL_SUBFOLDER = "Wan2.1-T2V-14B"
 HF_REPO = "RaphaelLiu/PusaV1"
@@ -45,40 +54,39 @@ def download_model_subset():
 
 
 @spaces.GPU
-def generate_video(prompt):
-    download_model_subset()
-
-    temp_output_dir = "/tmp/pusa_video_output"
-    os.makedirs(temp_output_dir, exist_ok=True)
-
-    command = [
-        "python", PUSA_SCRIPT_PATH,
-        "--prompt", prompt,
-        "--lora_path", FINAL_MODEL_PATH,
-        "--output_dir", temp_output_dir
-    ]
-
+def generate_video(prompt: str):
     try:
-        print("🚀 Running inference...")
-        subprocess.run(command, check=True)
+        # Load model manager
+        manager = ModelManager(base_model_dir=WAN_MODEL_DIR)
+        model = manager.load_model()
+
+        # Create video pipeline and apply LoRA
+        pipeline = WanVideoPusaPipeline(model=model)
+        pipeline.set_lora_adapters(LORA_PATH)
 
-        # Return first mp4 video found
-        for file in os.listdir(temp_output_dir):
-            if file.endswith(".mp4"):
-                return os.path.join(temp_output_dir, file)
+        # Generate video
+        result = pipeline(prompt=prompt)
 
-        return "❌ No video generated."
+        # Save video to a temporary file
+        tmp_dir = tempfile.mkdtemp()
+        video_path = os.path.join(tmp_dir, "output.mp4")
+        save_video(result, video_path)
 
-    except subprocess.CalledProcessError as e:
-        return f"❌ Inference failed: {str(e)}"
+        return video_path
 
+    except Exception as e:
+        print(f"[ERROR] {e}")
+        return None
 
+# Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("## 🧘‍♂️ PusaV1 Text-to-Video Generator (Wan2.1-T2V-14B)")
-    prompt_input = gr.Textbox(label="Enter your prompt", lines=4, placeholder="A coral reef full of colorful fish...")
-    generate_button = gr.Button("Generate Video")
+    gr.Markdown("## 🎥 PusaV1 Text-to-Video Generator")
+    gr.Markdown("Describe a scene and generate a short video using Wan2.1-T2V + Pusa LoRA!")
+
+    prompt_input = gr.Textbox(label="Enter Prompt", lines=4, placeholder="E.g. A coral reef full of colorful fish...")
+    generate_btn = gr.Button("Generate Video")
     video_output = gr.Video(label="Generated Video")
 
-    generate_button.click(fn=generate_video, inputs=prompt_input, outputs=video_output)
+    generate_btn.click(fn=generate_video, inputs=prompt_input, outputs=video_output)
 
 demo.launch()