tomasz-syn-grzegorza/ubuntu-bare-metal
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add stack Update, ComfyUI model manager, and slim ComfyUI stack.

Browse Source 
Server UI gains Update on stack cards, ComfyUI Models tab with workflow
scan and downloads, and centralized comfyui_config. Model catalog and
download scripts move from stacks/comfyui to server-ui so ComfyUI stays a
minimal Docker wrapper for easier image updates.
This commit is contained in:
tomasz-syn-grzegorza
2026-07-05 18:45:17 +00:00
parent 359afb3a59
commit 73e4fc005e
33 changed files with 3604 additions and 307 deletions
Download Patch File Download Diff File Expand all files Collapse all files
stacks/comfyui/.env.example
+6
View File
@@ -12,3 +12,9 @@ CUDA_VISIBLE_DEVICES=0
# Extra CLI args passed to ComfyUI (e.g. --fast)
CLI_ARGS=
# Hugging Face token for gated models (optional)
# HF_TOKEN=
# Civitai API token for civitai catalog entries (optional)
# CIVITAI_TOKEN=
stacks/comfyui/README.md
+40 -90
View File
@@ -1,134 +1,84 @@
# ComfyUI stack
[ComfyUI](https://github.com/comfyanonymous/ComfyUI) — grafowy interfejs do generowania obrazów (Stable Diffusion, Flux, …). Stack oparty na obrazie [`yanwk/comfyui-boot`](https://github.com/YanWenKun/ComfyUI-Docker).
Zamiast [Stability Matrix](https://github.com/LykosAI/StabilityMatrix) (GUI desktop) używamy ComfyUI w Dockerze — zgodnie z filozofią headless serwera i tutoriala [03b](../../manual-tutorial/03b-system-tools.md).
[ComfyUI](https://github.com/comfyanonymous/ComfyUI) w Dockerze — obraz upstream [`yanwk/comfyui-boot:cu126-slim`](https://github.com/YanWenKun/ComfyUI-Docker). Stack **nie modyfikuje** ComfyUI; to tylko compose + lifecycle. Katalog modeli, pobieranie i skan workflow są w [**Server UI**](../server-ui/README.md) (`:8091` → zakładka **ComfyUI Modele**).
## Porty
| Serwis | Port | URL |
|--------|------|-----|
| ComfyUI web UI | **8188** | `http://HOST:8188` |
| LocalAI (osobny stack) | 8070 | LLM / chat — **nie równolegle z dużym modelem SD** |
| Server UI (modele, stacki) | **8091** | `http://HOST:8091` |
## Jak to działa
```mermaid
flowchart LR
browser["Przeglądarka :8188"]
comfyui["Kontener comfyui"]
gpu["RTX 3090 Ti"]
models["/data/apps/comfyui/models"]
browser --> comfyui
comfyui --> gpu
comfyui --> models
```
| Element | Opis |
|---------|------|
| Obraz | `yanwk/comfyui-boot:cu126-slim` |
| Konfiguracja | `.env` + `docker-compose.yml` |
| Modele | `/data/apps/comfyui/models` (puste na start — pobierz ręcznie lub przez ComfyUI-Manager) |
| Pierwszy start | Kopiuje ComfyUI do `/data/apps/comfyui/storage/` (~kilka minut) |
## Struktura
## Struktura (repo)
```
stacks/comfyui/
├── README.md
├── docker-compose.yml
├── .env.example
├── .gitignore
└── scripts/
├── ensure-dirs.sh
├── pull.sh
└── start.sh
```
Na dysku `/data`:
## Dane na hoście (`/data/apps/comfyui/`)
```
/data/apps/comfyui/
├── storage/ # kopia ComfyUI z obrazu (pierwszy start)
├── models/ # checkpoints, LoRA, VAE, …
├── cache/
│ ├── hf-hub/
│ └── torch-hub/
├── input/
├── output/
├── custom_nodes/
└── workflows/
```
| Katalog | Zawartość |
|---------|-----------|
| `storage/` | Kopia ComfyUI z obrazu (pierwszy start) |
| `models/` | Checkpointy, LoRA, VAE, … |
| `custom_nodes/` | ComfyUI-Manager i wtyczki |
| `workflows/` | JSON workflow (bind → `user/default/workflows`) |
| `input/`, `output/`, `cache/` | Wejścia, wyniki, cache HF/torch |
## Workflow (bez modelu)
Update kontenera (`docker compose pull` + recreate) **nie usuwa** tych danych — wszystko to bind-mounty.
## Szybki start
```bash
cd stacks/comfyui
cp .env.example .env
./scripts/pull.sh
./scripts/start.sh
```
Weryfikacja:
Weryfikacja: `curl -f http://127.0.0.1:8188/`
## Polityka GPU
RTX 3090 Ti — jeden duży workload GPU naraz (LocalAI **lub** ComfyUI z dużym modelem). `start.sh` ostrzega, gdy `localai` działa. W Server UI: Stop/Start stacków.
## Update ComfyUI
Zalecane: Server UI → karta **ComfyUI** → **Update** (pull + recreate gdy działa).
CLI:
```bash
curl -s -o /dev/null -w '%{http_code}\n' http://127.0.0.1:8188/
# UI: http://<IP-serwera>:8188
./scripts/pull.sh # tylko obraz
./scripts/start.sh # start z nowym obrazem
```
## Zmienne `.env`
### Checklist po update obrazu
| Zmienna | Opis | Domyślnie |
|---------|------|-----------|
| `DATA_ROOT` | Mount dysku danych | `/data` |
| `COMFYUI_PORT` | Port na hoście | `8188` |
| `COMFYUI_IMAGE` | Obraz Docker | `yanwk/comfyui-boot:cu126-slim` |
| `CUDA_VISIBLE_DEVICES` | GPU | `0` |
| `CLI_ARGS` | Dodatkowe flagi ComfyUI | puste |
1. `curl -f http://127.0.0.1:8188/` — UI ComfyUI działa
2. Server UI → **ComfyUI Modele** → odśwież — skan workflow OK
3. Test pobrania modelu (katalog lub Custom URL) — job `completed`
4. Healthcheck fail → sprawdź, czy obraz ma `curl` na porcie 8188
5. Pusty skan workflow → zweryfikuj mount `workflows` → `user/default/workflows` w [docs yanwk](https://github.com/YanWenKun/ComfyUI-Docker)
## Polityka GPU (LocalAI ↔ ComfyUI)
## Modele i workflow
RTX 3090 Ti 24 GB — **jeden** duży workload GPU naraz.
- **Server UI** (zalecane): katalog, workflow scan, pobieranie — [`../server-ui/README.md`](../server-ui/README.md)
- **ComfyUI-Manager** w UI `:8188`
- **CLI** (Server UI scripts): `../server-ui/scripts/comfyui/list-models.sh`, `download-model.sh`
Przed startem ComfyUI z dużym modelem (SDXL, Flux):
Przykładowy workflow: [`../../manual-tutorial/examples/comfyui/example-sdxl-refiner.api.json`](../../manual-tutorial/examples/comfyui/example-sdxl-refiner.api.json)
```bash
cd ../localai
docker compose --profile localai stop localai
```
Skrypt `start.sh` ostrzega, gdy `localai` jest uruchomiony.
W Server UI (port 8091) → Stop/Start stack `localai` lub `comfyui` według potrzeb.
## Modele (później)
- **ComfyUI-Manager** w UI (custom node w obrazie yanwk) — pobieranie modeli i węzłów
- Ręcznie: pliki do `/data/apps/comfyui/models/checkpoints/` (lub odpowiednie podkatalogi)
Szacunki VRAM (przy zatrzymanym LocalAI):
| Model | VRAM (orientacyjnie) |
|-------|----------------------|
| SD 1.5 | ~4–6 GB |
| SDXL | ~8–12 GB |
| Flux | ~12–20 GB |
## Zarządzanie
```bash
docker compose --profile comfyui ps
docker compose --profile comfyui logs -f comfyui
docker compose --profile comfyui restart comfyui
docker compose --profile comfyui down
cp manual-tutorial/examples/comfyui/example-sdxl-refiner.api.json /data/apps/comfyui/workflows/
```
## Dokumentacja
- Tutorial: [manual-tutorial/07-comfyui-stack.md](../../manual-tutorial/07-comfyui-stack.md)
- Research Stability Matrix: [`coding-agent/STABILITYMATRIX-RESEARCH.md`](../../coding-agent/STABILITYMATRIX-RESEARCH.md)
- Wdrożenie: [`coding-agent/COMFYUI-DEPLOYMENT.md`](../../coding-agent/COMFYUI-DEPLOYMENT.md)
Upstream: [github.com/comfyanonymous/ComfyUI](https://github.com/comfyanonymous/ComfyUI)
- Wdrożenie: [coding-agent/COMFYUI-DEPLOYMENT.md](../../coding-agent/COMFYUI-DEPLOYMENT.md)
stacks/comfyui/compose.yaml
-1
View File
@@ -1 +0,0 @@
docker-compose.yml
stacks/comfyui/scripts/ensure-dirs.sh
+13 -1
View File
@@ -5,13 +5,25 @@ ensure_comfyui_dirs() {
local data_root="${1:-/data}"
mkdir -p \
"${data_root}/apps/comfyui/storage" \
"${data_root}/apps/comfyui/models" \
"${data_root}/apps/comfyui/models/checkpoints" \
"${data_root}/apps/comfyui/models/loras" \
"${data_root}/apps/comfyui/models/vae" \
"${data_root}/apps/comfyui/models/controlnet" \
"${data_root}/apps/comfyui/models/upscale_models" \
"${data_root}/apps/comfyui/models/clip" \
"${data_root}/apps/comfyui/models/unet" \
"${data_root}/apps/comfyui/cache/hf-hub" \
"${data_root}/apps/comfyui/cache/torch-hub" \
"${data_root}/apps/comfyui/input" \
"${data_root}/apps/comfyui/output" \
"${data_root}/apps/comfyui/custom_nodes" \
"${data_root}/apps/comfyui/workflows"
# Host downloads (Server UI / CLI) need write access; container (root) can still write.
local owner="${COMFYUI_DATA_OWNER:-${SUDO_USER:-}}"
if [[ "${EUID}" -eq 0 && -n "${owner}" ]]; then
chown -R "${owner}:${owner}" "${data_root}/apps/comfyui/models"
fi
}
if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then