For ComfyUI users who want their workflows inside a larger production: a timeline, storyboards, tools for clients and an assistant that writes missing steps, without giving up the ComfyUI graphs they already trust.
ComfyUI is an open-source node engine for image, video, audio and 3D models. It runs free on your own GPU, and Comfy Org also offers a hosted version, Comfy Cloud.
Kaitoi is a hosted creative platform where AI models, media tools and Python code sit in one graph, with a storyboard, a timeline and an assistant, Penny, that writes new nodes. It also connects to your own ComfyUI and turns a whole ComfyUI workflow into one Kaitoi node, so the two can work together rather than one replacing the other.
Checked against each product's own documentation on .
ComfyUI's core is open source under GPL-3.0 and free to run. Comfy Desktop installs it on Windows, macOS and Linux, it runs on NVIDIA, AMD, Intel and Apple Silicon GPUs, and once models are downloaded it works fully offline.
It runs a wide range of open models natively, across image, video, audio, 3D and Gaussian splats. Comfy says more than 5,000 community extensions add over 60,000 nodes, which ComfyUI-Manager and the Comfy Registry let you find, install and pin.
It gives precise control. Every parameter is a node you can see and change, only the parts of a graph that changed run again, and saved outputs carry the workflow that made them, so a result can be reproduced.
The hosted side keeps growing: Comfy Cloud, Partner Nodes for closed models, an API that deploys a workflow as an endpoint, App Mode for a simple form over a workflow, and Comfy Agent, a chat assistant that builds and runs workflows on Comfy Cloud.
Open source, coming soon. Kaitoi Studio is hosted and free to start, with no subscription, and Kaitoi Open Source, a distribution you run yourself, is announced as coming soon. Your own GPU joins through Kaitoi CLI, and Kaitoi Code runs on Ollama, LM Studio or any OpenAI-compatible server with no Kaitoi login; Studio itself needs a connection. There is no desktop app; Studio runs in the browser.
Open models through your own ComfyUI, and more. Kaitoi connects your ComfyUI through Kaitoi CLI, on your own computer or a GPU machine elsewhere, and on Studio wraps an exported workflow as one node, so the models and custom nodes installed in your ComfyUI keep working. Hosted providers add more than 1,400 generative models, 3D generation and a Gaussian splat viewer sit in the same graph, and nodes and templates publish to Kaitoi's Community.
Control you can see. Every step is a node with a declared interface, a run reuses the results of steps that did not change, and you can run one node or a selection and stop a run. A wrapped ComfyUI workflow keeps its parameters inside, with the ones you choose exposed as inputs. The project keeps the graph, with templates and backups to return to an earlier version, but generated files do not carry the graph that made them.
The hosted side is the default. Studio is hosted, with more than 1,400 generative models on credits at the provider's own price; a saved template becomes an API endpoint; a Station turns a workflow into a simple form or a custom app; and, on every plan, Penny builds, edits and runs graphs and writes missing nodes.
Beyond that, the graph is general purpose. Nodes are Python, and a node can carry its own HTML, CSS and JavaScript interface, WebAssembly included, anything from a small control to a full editor, so one graph can mix a ComfyUI workflow with file processing, code and hosted models.
Models that are not in ComfyUI can run on your hardware too, and that hardware does not have to be the computer you work at: Kaitoi CLI can run on a GPU machine elsewhere while you use Studio from a laptop. Kaitoi Code, the coding agent in Kaitoi CLI, turns a Hugging Face model or a GitHub repository into a Runtime recipe with declared inputs and outputs, tests it and, on the Studio plan, creates the node.
The graph is not the only view. The same project has Blocks, a simpler chain view of the same nodes, a storyboard, and a timeline with picture and sound tracks that renders video and exports OpenTimelineIO. When the work is for other people, a project becomes a Station at a stable link: a simple form by default, or a custom app built from a node's own interface. Stations run in the cloud, so a step that needs your ComfyUI stays in Studio.
| Question | ComfyUI | Kaitoi |
|---|---|---|
| Where the work runs Whose hardware runs the models. | On your own GPU, free; on Comfy Cloud's hosted GPUs; or a local graph can send single nodes to Comfy's GPUs. | Studio is hosted and runs in the browser, with hosted models from Kaitoi's providers on credits. Your own hardware joins through Kaitoi CLI, on the computer you work at or a GPU machine elsewhere: ComfyUI, Ollama, Blender or Runtime recipes. Steps on your own hardware use no credits. |
| Custom nodes Adding a step the product does not ship. | Python nodes with optional JavaScript interface code, from a large community registry. Locally they can install any Python dependency. | Python nodes, using packages from an approved list, that can carry their own HTML, CSS and JavaScript interface, WebAssembly included. Code that needs any other dependency runs on your machine as a Runtime recipe. Penny writes and edits nodes on every plan, editing node code by hand needs Studio or Enterprise, and nodes can be shared through Community. |
| Assistant An AI that builds or changes the workflow for you. | Comfy Agent builds, edits and runs workflows from chat on Comfy Cloud. Support for local ComfyUI and agent-made custom nodes is listed as coming soon. | Penny builds, edits and runs graphs, writes new custom nodes, edits storyboards and timelines, and works through ideas with you in Lab, on every plan. Penny has no approval mode before runs; Kaitoi Code, in the terminal, has Plan, Build and Crew modes. On every plan, outside agents build graphs and run approved work through Kaitoi's MCP server, and on Studio they write nodes. |
| ComfyUI workflows Using workflows built in ComfyUI. | Native. Workflows are saved as JSON, and generated files carry the workflow that made them. | Wraps an exported workflow as one node that runs on your connected ComfyUI, exposing only the inputs and outputs you choose. Studio plan. The project keeps the graph, but generated files do not carry it. |
| Editing a finished piece Cutting shots together and handing off to an editor. | Nodes trim, join and save video as MP4, MKV, WebM or MOV. We found no timeline editor and no EDL, XML or OTIO export in its public docs. | A storyboard and a multi-track timeline in the project. Renders MP4, WebM or MOV in ProRes 422 HQ, and exports and imports OpenTimelineIO. |
| Sharing with people who do not build A simple page someone else can run. | App Mode turns a workflow into a simple form. Share links work on Comfy Cloud. | Stations at a stable link, published as frozen versions: a simple form by default, or a full custom app built from a node's own HTML, CSS and JavaScript interface. Each visitor runs on their own account and pays their own costs. A Station that runs your own code, Penny-written nodes included, is invite only. Steps that need your ComfyUI or another local service cannot go in a Station. |
| API Calling a workflow from your own software. | Comfy API deploys a workflow as an autoscaling endpoint billed per GPU second. A local install also exposes an HTTP API. | A saved template becomes an asynchronous API endpoint. A REST API runs graphs from your own code, and AI clients build graphs and run work through Kaitoi's MCP server. |
Each cell comes from the product's own website, documentation or code repository. "Not publicly documented" means we found no official page that says so, not that the feature is missing. Plans and features change; check the official pages before you decide.
Kaitoi CLI turns an external Hugging Face repository into a runnable local recipe, exposes its parameters inside Kaitoi, and generates a runtime node that joins the graph like any other capability.
This film, from our Kaitoi Signals series, takes a model that is not in any product and puts it on the graph.
Read more in the documentation:
Local ComfyUI is free. Comfy Cloud is a paid subscription, with a few free runs for new users, and Partner Nodes and Comfy Agent use prepaid Comfy credits.
Free to start, with no subscription. Credits pay for hosted models, including Penny's, at the provider's own price with no markup, and for paid Community nodes; steps that call neither, and your own ComfyUI, cost nothing. Your own provider keys work on every plan and are billed by your provider. Studio adds included credits, team seats and hand-edited code.
Yes. Export the workflow from ComfyUI, and Kaitoi wraps it as one node that runs on your ComfyUI, with the inputs and outputs you choose. Your ComfyUI has to be reachable from Studio, and creating the node needs the Studio plan.
No. Hosted models run on credits without your own hardware. A GPU is optional, for connecting your ComfyUI, Ollama, Blender or Runtime recipes, and it can be in another machine from the one you use Studio on.
Not today. Kaitoi Studio is a hosted service, and a distribution you run yourself, Kaitoi Open Source, is announced as coming soon. ComfyUI is open source under GPL-3.0 now.
Yes. Comfy Agent builds, edits and runs workflows on Comfy Cloud, with an approval mode before runs. Comfy lists support for local ComfyUI as coming soon.