# Qwen Image Layered - Docker Deployment for HF Inference Endpoints

This directory contains a custom Docker-based deployment for `QwenImageLayeredPipeline`.

## Files
- `Dockerfile`: Custom container with all bleeding-edge dependencies.
- `app.py`: FastAPI server (HF-compatible API format).
- `handler.py`: Model loading and inference logic.
- `requirements.txt`: Python dependencies (all from git main).

## Deployment Steps

### 1. Build and Push Docker Image
```bash
# Login to Docker Hub (or another registry)
docker login

# Build the image
docker build -t yourusername/qwen-layered:latest .

# Push to registry
docker push yourusername/qwen-layered:latest
```

### 2. Create HF Inference Endpoint
1. Go to [HF Inference Endpoints](https://ui.endpoints.huggingface.co/)
2. Click **New Endpoint**
3. Select **Custom Container**
4. Enter your Docker image URL: `docker.io/yourusername/qwen-layered:latest`
5. Select GPU (A10G or better, 24GB+ VRAM)
6. Deploy

### 3. Usage
```bash
curl https://your-endpoint.endpoints.huggingface.cloud \
  -X POST \
  -d '{"inputs": {"prompt": "A cute cat"}}' \
  -H "Authorization: Bearer hf_..." \
  -H "Content-Type: application/json"
```

## Local Testing
```bash
docker build -t qwen-test .
docker run --gpus all -p 8080:8080 qwen-test
# Then: curl http://localhost:8080/health
```