Commands — NUILab AI Game Art

Session

ssh f-nui-zeph-01 comfy-up                        # evict residents, start ComfyUI
ssh -N -L 8188:127.0.0.1:8188 f-nui-zeph-01       # tunnel (leave running)
ssh f-nui-zeph-01 comfy-down                      # stop, restore residents, verify

Then browse to http://127.0.0.1:8188 (opens in new tab) .

Prove it still works

After an update, or whenever you suspect the stack, generate something real rather than trusting that the services started:

~/projects/comfy-venv/bin/python ~/scripts/comfy-smoke.py sdxl    # classic lane, + pixel LoRA
~/projects/comfy-venv/bin/python ~/scripts/comfy-smoke.py klein   # primary lane
~/projects/comfy-venv/bin/python ~/scripts/comfy-smoke.py pixel   # generate -> grid detect -> quantize

Each queues a real workflow, waits for it, and prints the output filename, elapsed time, and peak VRAM. Non-zero exit on any failure, so it works as a post-install check. Requires comfy-up first.

Status

nvidia-smi --query-gpu=memory.used,memory.total --format=csv        # total VRAM in use
nvidia-smi --query-compute-apps=pid,used_memory,process_name --format=csv   # per-process
ollama ps                                                          # what the LLM holds
systemctl --user is-active scholar-api.service                     # is the RAG API up
ss -lptnH 'sport = :8188'                                          # who owns the ComfyUI port

Logs

tail -f ~/projects/logs/comfyui.log      # ComfyUI stdout/stderr
journalctl --user -u scholar-api -n 40   # RAG API, if a restore failed

Environment

~/projects/comfy-venv/bin/python -c "import torch; print(torch.__version__, torch.cuda.get_arch_list())"
curl -s http://127.0.0.1:8188/system_stats | python3 -m json.tool

Updating

cd ~/projects/comfyui && git pull
uv pip install --python ~/projects/comfy-venv/bin/python -r requirements.txt
~/projects/comfy-venv/bin/python -c "import torch; print(torch.__version__)"   # confirm still +cu130

The last line is not optional. An upgrade that pulls a generic PyPI torch over the CUDA build is the most common way this stack breaks, and it does not announce itself — see Troubleshooting .

Paths

Path	What
`~/projects/comfyui`	The ComfyUI checkout — disposable
`~/projects/comfy-venv`	The Python 3.12 environment
`~/models/`	All weights, outside the checkout
`~/projects/logs/`	ComfyUI and install logs
`~/scripts/comfy-{up,down}.sh`	The GPU arbitration scripts

Source: content/reference/commands.md · maintained in the nuilab-aigaming repository.