services:
  rmbg:
    build: .
    image: rmbg-as-a-service:latest
    container_name: rmbg-as-a-service
    ports:
      - "${PORT:-8000}:8000"
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
      # Default variant/resolution; both are also selectable per request.
      - BIREFNET_MODEL=${BIREFNET_MODEL:-general}
      - BIREFNET_RESOLUTION=${BIREFNET_RESOLUTION:-1024}
    # Use the nvidia-container-runtime for GPU acceleration.
    runtime: nvidia
    volumes:
      # Persist downloaded BiRefNet weights across container restarts.
      - hf-cache:/app/hf_cache
    healthcheck:
      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"]
      interval: 15s
      timeout: 5s
      retries: 30
      start_period: 180s
    restart: unless-stopped

volumes:
  hf-cache: