An infinite canvas where the AI agent does the editing, not just the generating

What it does

BeefTV is a desktop app (macOS and Windows) for AI-assisted video and media creation, built with React, Go, and Wails. You work on a free canvas where nodes and connections represent materials, context, and generation flows — then generate text, images, video, and audio from prompts or reference material, and keep processing the results (crop, annotate, inpaint, split, remix). The point is that a generation isn’t a dead end: everything stays editable, reusable, and iterable.

The interesting bit

The architecture decouples viewport rendering, node loading, media preview, and generation tasks, so the canvas stays responsive as projects grow — a claim many node editors make and few deliver, but at least it’s stated as a design goal. Also notable: it’s provider-agnostic, letting you configure your own model channels for text, image, video, and audio instead of locking you into one vendor.

Key highlights

• Local-first workspace: projects, canvases, assets, and tasks are stored and migratable locally
• Full creation chain on one canvas: generate, organize, process, iterate
• Agent-driven AIGC workflow — the AI participates in the process, not just the output
• Includes a project library, personal asset library, async tasks, and model channels
• MIT-licensed, TypeScript/Go codebase, ~1k stars

Caveats

• The README is mostly in Chinese; English speakers will lean on the demo video and code
• “High-performance” is asserted, not benchmarked — no numbers provided
• You must bring your own model channels; it’s a workspace, not a bundled AI service

Verdict

Worth a look if you do AI media generation and want a canvas that keeps the whole messy iterative process organized. Skip it if you expected a turnkey video editor with models included.