How it works
Three parties: the Consumer builds workflows in ComfyUI, a Provider lends an idle GPU, and GpuRent matches the two, moves the files and keeps the books. Both apps only connect out to GpuRent, so nobody opens ports or shares an IP address.
Creating with ComfyUI
Install the plugin.
One command finds Comfy Desktop and adds the GpuRent nodes.
One command finds Comfy Desktop and adds the GpuRent nodes.
Sign in with a code.
The GpuRent sidebar shows a code; confirm it on this website.
The GpuRent sidebar shows a code; confirm it on this website.
Queue on GpuRent.
Open a template or your own workflow and press the cloud button (Alt+Shift+G).
Open a template or your own workflow and press the cloud button (Alt+Shift+G).
Watch it render.
Live previews and progress on the node, then the result on the canvas and in My jobs.
Live previews and progress on the node, then the result on the canvas and in My jobs.
Providing a GPU
Install the Provider app.
It sets up a sandboxed ComfyUI runtime in Docker (WSL2).
It sets up a sandboxed ComfyUI runtime in Docker (WSL2).
Pick a disk budget and models.
GpuRent fetches more models on demand within your budget.
GpuRent fetches more models on demand within your budget.
Sign in with a code.
The machine appears under My GPUs.
The machine appears under My GPUs.
Leave it on.
Pause from the tray when you need the GPU. Running jobs move to another machine.
Pause from the tray when you need the GPU. Running jobs move to another machine.
What runs where
| Where | What happens |
|---|---|
| Your ComfyUI | Builds the graph, uploads input images, shows previews and results. Models are never needed locally. |
| GpuRent | Checks the workflow against the node allowlist and the model catalog, estimates the work, picks a GPU with the model warm, and stores inputs and outputs briefly. |
| Provider GPU | Runs the workflow in a pinned ComfyUI container, streams progress and previews, and uploads the outputs. |