ComfyUI in the cloud, on your own RunPod: one click, no idle GPU bill
Running ComfyUI in the cloud usually means renting a pod, waiting for it to boot, installing nodes and models, and paying for every minute it stays on — even while you think. Onyx.AI runs its ComfyUI workflows on serverless GPUs in your own RunPod account instead: a GPU starts for each job and stops after it, the models are stored once, and you generate from a simple form on the website.
Two ways to run ComfyUI on RunPod
| A pod with a ComfyUI template | Serverless, set up by Onyx.AI | |
|---|---|---|
| You get | The full ComfyUI interface | A form per workflow, results in your library |
| GPU billed | Every minute the pod is on | Only while a job runs (plus a warm-up delay you choose) |
| Models | Downloaded on the pod or a volume you manage | Stored once on a network volume, only for the workflows you add |
| Setup | Nodes, models and versions to keep in line | Created for you with your RunPod API key |
| Best for | Building and tweaking workflows | Running proven workflows, again and again |
Both have their place. Pack owners get the workflows and the custom nodes to open in ComfyUI on their own PC or on a pod; the Serverless GPU module is for the days you just want the result.
How the serverless setup works
- Add your RunPod API key. It stays in your browser and is sent to RunPod through Onyx.AI only for the call that needs it — never saved on our server.
- Pick the workflows you want. Only their models are stored, on one network volume in your account. Storage is billed by RunPod; a workflow you do not add costs nothing.
- Generate from the website. Each workflow has its own form. The job runs on your endpoint, and the result is saved in your Onyx.AI library.
- Pay for GPU time only while you work. Endpoints keep zero workers running when nothing is asked. You choose how long a GPU stays awake after a job: longer means no cold start for the next one, and RunPod bills that waiting time too.
The workflow itself is sent to your worker through a one-time ticket, not through RunPod's job input, so it does not end up in plain view in your RunPod console.
The workflows
- Krea 2 — amateur phone-photo look, 4K natural, image to image, and carousels with the same face slide after slide.
- MiniMax H3 — motion control and text or image to video, with original and new audio kept apart.
- Wan Animate and SCAIL 2 — motion transfer from a source video to your character.
- Realism refiner and SDXL inpainting — for the last ten percent.
All of them are included with the Serverless GPU module. For bigger batches without a GPU of your own, AI motion control also runs on API providers such as Kie.ai or WaveSpeed.
ComfyUI cloud services vs your own RunPod
Hosted ComfyUI services sell GPU time or credits at their price, on their machines. Here the GPUs, the storage and the bill are RunPod's, in your name, at RunPod's price — Onyx.AI provides the workflows, the setup and the interface, as a monthly module. If you stop the module, the RunPod account and what is in it stay yours.
What it does — and what it does not
What it does
- Sets up serverless ComfyUI endpoints in your RunPod account
- Stores only the models of the workflows you add
- Runs the Onyx.AI workflows from a form, results in your library
- Keeps your RunPod key out of our server
What it does not
- Resell GPU time or add a markup to RunPod's price
- Give you the full ComfyUI canvas in the cloud (use a pod for that)
- Keep a GPU running when nobody generates, unless you ask it to
- Speak for RunPod — Onyx.AI is independent
Common questions
How do I run ComfyUI on RunPod? Either a pod with a ComfyUI template, for the full interface, or serverless endpoints that start a GPU per job — which is what Onyx.AI sets up for you.
Do I pay for the GPU when I am not generating? No GPU time: zero workers run while nothing is asked. Storage, and any warm-up time you choose, are billed by RunPod.
Where is my RunPod key kept? In your browser, sent to RunPod only for the call that needs it, never saved on our server.
Can I still open the workflows in ComfyUI? Yes, with a pack: the workflows and the custom nodes are yours to run on your own PC or a pod.
More in the FAQ — billing, cancellation, privacy and what happens to your files.
Generate on your own GPUs, without the GPU
Every pack opens the whole platform free to start with — 15 days, 1 month or 2 months depending on the pack — Serverless GPU module included, with the workflows and custom nodes yours to keep.