40 lines
1.4 KiB
Markdown
40 lines
1.4 KiB
Markdown
# modules/nixos/server/services/comfyui/
|
|
|
|
ComfyUI (Stable Diffusion workflow UI) running in Docker with GPU acceleration. Supports both
|
|
**ROCm** (AMD) and **CUDA** (NVIDIA) via a `mode` switch, with shared options/assertions in
|
|
`shared.nix`.
|
|
|
|
## Files
|
|
|
|
```
|
|
comfyui/
|
|
├── default.nix # imports shared + rocm + rocm-docker + cuda
|
|
├── shared.nix # skyg.nixos.server.services.comfyui.* — options, firewall, assertions
|
|
├── rocm.nix # ROCm (AMD) path
|
|
├── rocm-docker.nix # ROCm container/image wiring
|
|
├── cuda.nix # CUDA (NVIDIA) path
|
|
├── rocm.Containerfile# ROCm image definition
|
|
└── init.sh # Container init script
|
|
```
|
|
|
|
## Behaviour
|
|
|
|
- `mode` selects `"rocm"` (default) or `"cuda"`. Web UI on `port` (default 8188),
|
|
`openFirewall` defaults to **true**.
|
|
- `shared.nix` asserts that Docker, `hardware.graphics`, and `hardware.enableAllFirmware` are
|
|
all enabled — ComfyUI needs the GPU passed through to the container.
|
|
|
|
## Option Namespace
|
|
|
|
```
|
|
skyg.nixos.server.services.comfyui.enable
|
|
skyg.nixos.server.services.comfyui.mode # "rocm" | "cuda"
|
|
skyg.nixos.server.services.comfyui.port # default 8188
|
|
skyg.nixos.server.services.comfyui.openFirewall # default true
|
|
```
|
|
|
|
## Conventions
|
|
|
|
- Pick the GPU `mode` matching the host hardware; the unused path stays inert.
|
|
- This module requires Docker (not Podman) and the relevant `hardware/*` GPU module enabled.
|