Install ComfyUI in one click

Press Install on ComfyUI’s card and it arrives with PyTorch and a working CUDA build. Save a snapshot, and every GPU machine you start already has it.

Installing ComfyUI is easy the first time. Installing it on a fresh cloud GPU every week is not: the right PyTorch for the card, a CUDA build that matches the driver, the custom nodes, the wheel that will not build.

In Hinode you install it once, on your workspace, and save it.

Ten apps, one click each

Each card says how much it downloads before you commit. What an app needs comes with it: install ComfyUI and PyTorch installs first, and ComfyUI’s card counts PyTorch in its size. A few minutes later it is done, and you can keep working while it runs.

Save it once

A machine starts from your newest snapshot. So when ComfyUI is installed, press Save on the workspace screen and add a line like “ComfyUI and the new LoRA nodes”. The workspace screen lists anything installed since the last save, so nothing is left behind by accident.

The five newest snapshots are kept. If a new custom node breaks a workflow, attach a machine from the snapshot before it.

Open ComfyUI on a GPU

Set up in your workspace. Render on a machine. Attach a machine with the VRAM the workflow needs, and open ComfyUI from its Connect menu, or in your own browser:

The tab opens on your computer through a private tunnel, so ComfyUI is never open to the internet.

Your own custom nodes

It is your computer. Clone custom nodes, pip install what they need, add models from Civitai or Hugging Face. None of it is second class. Put checkpoints and LoRAs in the shared folder, and every machine sees them without downloading them again.