Update and optimize for CPU.

Dockerfile

@@ -0,0 +1,31 @@
+# Dockerfile for MovieGen Demo (CPU-only by default)
+# To enable GPU, use a CUDA base image and install requirements-gpu.txt
+
+FROM python:3.12
+
+WORKDIR /app
+
+# System dependencies
+RUN apt-get update && apt-get install -y \
+    ffmpeg \
+    git \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy code
+COPY . /app
+
+# Install Python dependencies (CPU by default)
+RUN pip install --upgrade pip && \
+    pip install -r requirements.txt
+
+# Detect if running in a CUDA environment and install GPU dependencies if so
+RUN if python -c "import torch; print(torch.cuda.is_available())" | grep -q True; then \
+    pip install -r requirements-gpu.txt; \
+    echo 'Installed GPU dependencies.'; \
+else \
+    echo 'Running in CPU-only mode.'; \
+fi
+
+EXPOSE 7860
+
+CMD ["python", "app.py"]

HARDWARE.md

@@ -0,0 +1,32 @@
+# Hardware Requirements for MovieGen Demo
+
+## Minimum Requirements
+- **CPU:** Modern x86_64 CPU (8+ cores recommended)
+- **RAM:** 16 GB (32 GB recommended for larger resolutions)
+- **Disk:** At least 10 GB free space for model checkpoints and cache
+
+## Recommended (GPU) Requirements
+- **GPU:** NVIDIA GPU with CUDA support (Ampere or newer recommended)
+- **GPU Memory:** 16 GB VRAM minimum (24 GB+ for high resolutions)
+- **CUDA:** CUDA 11.7+ (if using GPU)
+- **Driver:** Latest NVIDIA driver compatible with your CUDA version
+
+## Notes
+- The app can run on CPU, but generation will be much slower compared to GPU.
+- For best performance and high-resolution video generation, a high-end GPU is strongly recommended.
+- Ensure sufficient disk space for temporary video and model files.
+- If running on CPU, expect longer processing times and possible memory limitations for large models or high resolutions.
+
+## How to Select Device
+- Use the `--device` argument to select `cpu` or `cuda` (GPU) at launch:
+  ```bash
+  python app.py --device cuda  # for GPU
+  python app.py --device cpu   # for CPU
+  ```
+
+## Troubleshooting
+- If you encounter out-of-memory errors, try reducing the resolution or batch size.
+- If no compatible GPU is found, the app will default to CPU mode.
+
+---
+For more details, see the main README or contact the maintainers.

README.md

@@ -12,3 +12,32 @@ short_description: A demo space for MovieGen1.1.
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+## Hardware Requirements
+See [HARDWARE.md](./HARDWARE.md) for detailed hardware requirements and recommendations for running this app efficiently on CPU or GPU.
+
+## Running with Docker
+
+You can run this app in a containerized environment using Docker. The provided `Dockerfile` will automatically detect if a GPU is available and install GPU-specific dependencies (like `flash-attn`) if needed.
+
+### 1. Build the Docker image
+```bash
+docker build -t moviegen-demo .
+```
+
+### 2. Run the container
+```bash
+docker run --rm -p 7860:7860 moviegen-demo
+```
+
+- The app will be available at http://localhost:7860
+- By default, the container runs in CPU mode. If a compatible GPU is available and Docker is configured for GPU access (e.g., with `--gpus all`), GPU dependencies will be installed and used automatically.
+
+#### To run with GPU (if available):
+```bash
+docker run --rm --gpus all -p 7860:7860 moviegen-demo
+```
+
+> **Note:**
+> - For GPU support, you must have the [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html) installed and a CUDA-compatible GPU.
+> - See [HARDWARE.md](./HARDWARE.md) for more details on requirements.

app.py

@@ -6,6 +6,7 @@ import sys
 import warnings
 
 import gradio as gr
+import torch
 
 warnings.filterwarnings('ignore')
 
@@ -69,10 +70,17 @@ def gradio_interface():
                     <div style="text-align: center; font-size: 16px; font-weight: normal; margin-bottom: 20px;">
                         Wan: Open and Advanced Large-Scale Video Generative Models.
                     </div>
+                    <div style='text-align: center; color: #1976d2; font-size: 18px; margin-bottom: 16px;'><b>Upvote this space if you want faster generations!<br>We are submitting for a GPU grant after 100 upvotes 🚀</b></div>
                     """)
 
         with gr.Row():
             with gr.Column():
+                device_choice = gr.Radio(
+                    choices=["cuda", "cpu"],
+                    value="cuda" if torch.cuda.is_available() else "cpu",
+                    label="Select Device (GPU or CPU)",
+                    info="Choose 'cuda' for GPU (faster, if available) or 'cpu' for CPU mode."
+                )
                 txt2vid_prompt = gr.Textbox(
                     label="Prompt",
                     placeholder="Describe the video you want to generate",
@@ -135,12 +143,29 @@ def gradio_interface():
             inputs=[txt2vid_prompt, tar_lang],
             outputs=[txt2vid_prompt])
 
+        def run_generation_with_device(device, *args):
+            global wan_t2v, prompt_expander
+            # Re-initialize models if device changes
+            import wan
+            from wan.configs import WAN_CONFIGS
+            from wan.utils.prompt_extend import DashScopePromptExpander, QwenPromptExpander
+            cfg = WAN_CONFIGS['t2v-14B']
+            prompt_expander = QwenPromptExpander(model_name=None, is_vl=False, device=device)
+            wan_t2v = wan.WanT2V(
+                config=cfg,
+                checkpoint_dir="cache",
+                device_id=0 if device == "cuda" else -1,
+                rank=0,
+                t5_fsdp=False,
+                dit_fsdp=False,
+                use_usp=False,
+                device=device
+            )
+            return t2v_generation(*args)
+
         run_t2v_button.click(
-            fn=t2v_generation,
-            inputs=[
-                txt2vid_prompt, resolution, sd_steps, guide_scale, shift_scale,
-                seed, n_prompt
-            ],
+            fn=run_generation_with_device,
+            inputs=[device_choice, txt2vid_prompt, resolution, sd_steps, guide_scale, shift_scale, seed, n_prompt],
             outputs=[result_gallery],
         )
 
@@ -167,7 +192,12 @@ def _parse_args():
         type=str,
         default=None,
         help="The prompt extend model to use.")
-
+    parser.add_argument(
+        "--device",
+        type=str,
+        default="cuda" if torch.cuda.is_available() else "cpu",
+        choices=["cpu", "cuda"],
+        help="Device to run the model on (cpu or cuda). Default: cuda if available, else cpu.")
     args = parser.parse_args()
 
     return args
@@ -182,7 +212,7 @@ if __name__ == '__main__':
             model_name=args.prompt_extend_model, is_vl=False)
     elif args.prompt_extend_method == "local_qwen":
         prompt_expander = QwenPromptExpander(
-            model_name=args.prompt_extend_model, is_vl=False, device=0)
+            model_name=args.prompt_extend_model, is_vl=False, device=args.device)
     else:
         raise NotImplementedError(
             f"Unsupport prompt_extend_method: {args.prompt_extend_method}")
@@ -193,11 +223,12 @@ if __name__ == '__main__':
     wan_t2v = wan.WanT2V(
         config=cfg,
         checkpoint_dir=args.ckpt_dir,
-        device_id=0,
+        device_id=0 if args.device == "cuda" else -1,
         rank=0,
         t5_fsdp=False,
         dit_fsdp=False,
         use_usp=False,
+        device=args.device
     )
     print("done", flush=True)
 

requirements-gpu.txt

@@ -0,0 +1,3 @@
+# Additional requirements for GPU environments
+# Use this file if you have a CUDA-enabled GPU and want to use flash-attn
+flash_attn

requirements.txt

@@ -10,7 +10,9 @@ imageio
 easydict
 ftfy
 imageio-ffmpeg
-flash_attn
+# flash_attn is only needed for GPU environments
+# Use requirements-gpu.txt for GPU installs
+# flash_attn
 gradio>=5.0.0
 numpy>=1.23.5,<2
 xfuser