BeefTV is a local-first node canvas for AI video work, with text, image, video and audio generation on one project surface

Most tools for AI video work keep the project state inside a vendor dashboard, which means the prompts, the generated takes and the final cut all live on somebody else's server. BeefTV, an open-source desktop app at glanderness/BeefTV, keeps them on the operator's own disk and treats the AI calls as nodes on a canvas rather than buttons on a page.

The MIT-licensed project reached 798 stars and 163 forks ten days after its first public commit on 2026-09-24. Desktop builds exist for macOS and Windows, published to the GitHub releases page. The project is maintained from the @beefnoode X account.

One canvas, four kinds of output

The project describes four jobs that all happen on the same surface. Generate text, image, video and audio from prompts or reference files. Organise them with nodes and edges that record how one output feeds the next. Process the results with crop, mask, local repaint, split, reference and merge. Iterate by keeping every version and reusing earlier assets.

The point is that a video project is rarely one generation. It is a prompt that produces an image, an image that seeds a video, a video that needs a sound bed, and a cut that references all three. BeefTV holds those steps as explicit nodes rather than losing them in a chat history, so a later run can replace one upstream node and watch everything downstream regenerate.

What the project runs on

Projects, canvases, assets and tasks all share one workspace that saves to local storage and can be moved between machines. The desktop app is built with React, Go and Wails. The author describes the project as not tied to any one model provider: a user adds model profiles for text, image, video and audio separately, and the node that calls each one reads from the configured profile.

The architecture is described as decoupled: the viewport rendering, node loading, media preview and generation tasks run as separate concerns, so a canvas with many nodes stays responsive. The project page notes that the design target is a low resource footprint and a low setup cost, with abilities loaded only when a node calls for them.

Most of the documentation is in Chinese, including the quickstart (QUICKSTART.md) and the feature list under docs/content/docs/overview/features.mdx. The README repeats the key sections in English through the badge links and the Why BeefTV table. A fresh install needs no account, since the keys are the user's own.

The project is still at 1.5.5, with the README stating that defaults, file formats and interfaces can still move between releases. For a creator who wants an AI video pipeline on their own machine today, with their own keys and their own files, that is already enough to try.

Source

Primary source: BeefTV on GitHub and the project site at beeftv.app.