ComfyUI hosting

Node-based image and video generation — waiting on GPU capacity.

Speech, image & videoGPL-3.0on request

In short

ComfyUI is in our catalogue but not yet orderable: it needs GPU capacity, and we would rather list it honestly than sell a CPU plan that disappoints. Tell us what you need and we will tell you where it stands.

Node graphReproducible workflowsVideo modelsGPU required

ComfyUI is a node-based interface and backend for diffusion models, created by comfyanonymous and now developed by Comfy Org under the GPL-3.0 licence. Every stage of a generation pipeline, from loading a checkpoint to encoding the prompt, sampling and decoding, is a node you wire together. The finished graph is a reproducible workflow: saved as JSON, embedded in every PNG it produces, and loaded back by dragging the image onto the canvas, seeds and all.

It supports the current open model families for images and video, including Stable Diffusion 1.5 and SDXL, SD3.5, Flux, Wan 2.x, HunyuanVideo and LTX-Video, and a large ecosystem of custom nodes installed through ComfyUI-Manager. It also has a local API for running workflows from other applications. What it does not have is a good CPU story: a single image that takes seconds on a graphics card takes minutes without one, which is why it is on our list but not yet for sale.

What ComfyUI needs

A ComfyUI deployment is a single Python process serving the web UI and API on port 8188, with model checkpoints, LoRAs and custom nodes on local disk; there is no database, but model folders quickly reach tens or hundreds of gigabytes.

What we configure for you

--listen and --port 8188 behind TLS, since ComfyUI binds to 127.0.0.1 by default
extra_model_paths.yaml so checkpoints, LoRAs and VAEs live on a separate volume
--enable-manager for installing and updating custom nodes through ComfyUI-Manager
--multi-user for per-user storage of workflows and settings
--disable-api-nodes if you want everything to stay offline, with no paid partner nodes

Which plan fits

No plans are orderable yet. A useful instance needs an NVIDIA GPU with at least 12 GB of VRAM for SDXL-class work and 24 GB for Flux and current video models, plus fast storage for model files. Our CPU plans are not a substitute.

The honest limits

ComfyUI on CPU is technically possible and practically unusable for anything beyond testing a graph. Custom nodes are third-party Python code that runs with full access to the instance, so install them from sources you trust.

What people run ComfyUI for

Reproducible image pipelines

Share a workflow file and get the same result on another machine with the same models and seed.

Video generation

Chain text-to-video or image-to-video models such as Wan or LTX-Video into a single graph.

Image generation behind an API

Queue a saved workflow over the local API from another application or from Open WebUI.

Batch editing and upscaling

Run the same inpainting, upscaling or background-removal graph across a folder of images.

Licence and version policy

Item	Position
Upstream licence	GPL-3.0
Sold as managed hosting	No
Tenancy	Single tenant — one customer per instance
Patching	Critical advisories within 24 hours, high severity within 72

Our full position on licences, including the applications we decline to sell, is on the licensing page.

Reading on ComfyUI

ComparisonComfyUI alternatives

Works well with ComfyUI

More Speech, image & video we host

Questions

ComfyUI hosting: your questions

When will ComfyUI hosting be available?

When we can offer GPU plans at a price we are prepared to stand behind. We will not sell it on CPU: ComfyUI has a --cpu mode, but its own help text describes it as slow, and for diffusion models slow means minutes per image. Contact us if you would like to hear when GPU plans open.

How much GPU memory does ComfyUI need?

It depends on the model rather than on ComfyUI. SD 1.5 runs on modest cards and SDXL is comfortable around 8-12 GB of VRAM. Flux and the larger video models are much happier with 16-24 GB, although ComfyUI’s memory management and quantised model files let some of them run, more slowly, on smaller cards.

Are workflows portable between installations?

Yes. A workflow is a JSON file, and ComfyUI embeds it in the images it saves, so dragging a PNG onto the canvas restores the whole graph including seeds. The other machine needs the same model files and any custom nodes the workflow uses, which ComfyUI-Manager can detect and install.

Can other applications call ComfyUI?

Yes. ComfyUI exposes a local HTTP and WebSocket API for queueing workflows and fetching results, which is what integrations such as Open WebUI’s image generation use. A workflow exported in API format can be submitted from any language that can make an HTTP request.

What does the GPL-3.0 licence mean for my images?

The licence covers ComfyUI’s code, not what you make with it. Your images and videos are governed by the licences of the models you use, which vary widely: some checkpoints allow commercial use and some do not, so check each model card before you rely on its output commercially.

Tell us what you need

If this is the app you need, say so — demand is how things move up the list.

Browse the catalogue Talk to a human