ComfyUI

Run ComfyUI for image and video generation, and use its web UI from your browser.

GPU: 16 GB VRAM for SDXL, 24 GB or more for larger image and video models.

1. Install

curl -LsSf https://astral.sh/uv/install.sh | sh
source $HOME/.local/bin/env

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
uv venv --python 3.12 --seed
source .venv/bin/activate
uv pip install torch torchvision torchaudio --torch-backend=auto
uv pip install -r requirements.txt

--torch-backend=auto selects the PyTorch build that matches the installed driver.

2. Add a model

Checkpoints go in models/checkpoints. Example with SDXL base:

uv pip install huggingface_hub
hf download stabilityai/stable-diffusion-xl-base-1.0 sd_xl_base_1.0.safetensors \
  --local-dir models/checkpoints

Other model types have their own folders under models/ (loras, vae, controlnet, …).

Upload your own files from your computer:

rsync -avz ./my-model.safetensors ubuntu@203.0.113.42:~/ComfyUI/models/checkpoints/

3. Start

Run it inside tmux:

tmux new -s comfy
cd ~/ComfyUI && source .venv/bin/activate
python main.py --port 8188

Detach with Ctrl+b d. ComfyUI listens on 127.0.0.1:8188.

4. Open the UI

On your computer:

ssh -N -L 8188:localhost:8188 ubuntu@203.0.113.42

Open http://localhost:8188. The UI is reachable only through your tunnel.

5. Get your images

Outputs are written to ~/ComfyUI/output:

rsync -avz ubuntu@203.0.113.42:~/ComfyUI/output/ ./comfy-output/

Export your workflows from the UI (Workflow → Export) before the end. Everything on the machine is deleted with it.

Reference: ComfyUI on GitHub.