76 lines
2.6 KiB
Markdown
76 lines
2.6 KiB
Markdown
# ComfyUI for Framework Desktop AI MAX
|
|
|
|
A containerized setup for running [ComfyUI](https://github.com/comfyanonymous/ComfyUI) on the Framework Desktop AI MAX with AMD Ryzen AI MAX (Strix Halo) and integrated Radeon graphics.
|
|
|
|
## Requirements
|
|
|
|
- Framework Desktop AI MAX
|
|
- Docker with Compose plugin
|
|
- ROCm-compatible kernel and drivers
|
|
|
|
## Quick Start
|
|
|
|
1. **Build and start the container:**
|
|
|
|
```bash
|
|
docker compose up -d --build
|
|
```
|
|
|
|
2. **Access the UI:**
|
|
|
|
Open your browser and navigate to `http://localhost:8188`
|
|
|
|
3. **View logs:**
|
|
|
|
```bash
|
|
docker compose logs -f comfyui
|
|
```
|
|
|
|
4. **Stop the container:**
|
|
|
|
```bash
|
|
docker compose down
|
|
```
|
|
|
|
## Directory Structure
|
|
|
|
- `models/` — Place your model files here (checkpoints, VAE, LoRAs, etc.)
|
|
- `output/` — Generated images and videos are saved here
|
|
|
|
## Environment Variables
|
|
|
|
The Containerfile sets several ROCm-specific environment variables for Strix Halo compatibility:
|
|
|
|
| Variable | Value | Description |
|
|
| -------------------------- | --------------------------- | -------------------------------------------------------- |
|
|
| `HSA_OVERRIDE_GFX_VERSION` | `11.5.1` | Forces ROCm to recognize gfx1151 (Strix Halo) |
|
|
| `HSA_ENABLE_SDMA` | `0` | Disables System DMA engine to avoid compatibility issues |
|
|
| `ROCBLAS_USE_HIPBLASLT` | `1` | Enables optimized BLAS library for better performance |
|
|
| `PYTORCH_HIP_ALLOC_CONF` | `expandable_segments:False` | Uses simpler memory allocation for stability |
|
|
|
|
## Included Custom Nodes
|
|
|
|
- [ComfyUI-Manager](https://github.com/ltdrdata/ComfyUI-Manager) — Node management and installation
|
|
- [ComfyUI-WanVideoWrapper](https://github.com/kijai/ComfyUI-WanVideoWrapper) — WAN 2.1 video generation
|
|
- [ComfyUI-VideoHelperSuite](https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite) — Video utilities
|
|
|
|
## Troubleshooting
|
|
|
|
### Segmentation faults during model loading
|
|
|
|
Ensure the ROCm base image version matches the PyTorch ROCm wheels. Mismatched versions cause ABI incompatibilities.
|
|
|
|
### GPU not detected
|
|
|
|
Verify that `/dev/kfd` and `/dev/dri` are accessible and your user is in the `video` and `render` groups:
|
|
|
|
```bash
|
|
sudo usermod -aG video,render $USER
|
|
```
|
|
|
|
Log out and back in for the group changes to take effect.
|
|
|
|
### Performance issues
|
|
|
|
If using `HIP_LAUNCH_BLOCKING=1` for debugging, remember to remove it for normal use — it forces synchronous execution and significantly reduces performance.
|