About this project

Timeline Studio is an open-source, local-first video editor that runs in the browser. It provides a CapCut-style multi-track timeline for visuals, picture-in-picture overlays, captions, stickers, voiceover, separated source audio, and music. Human users and AI agents can work on the same editable project, with the portable .timeline file kept as the source of truth rather than producing only an opaque rendered video. Editing features include canvas selection and transforms, proportional resizing, rotation, masks, filters, effects, animation, speed changes, explicit keyframes, snapping, alignment guides, split/duplicate/delete operations, timeline zoom, and undo/redo. Desktop color grading supports keyframeable temperature, tint, saturation, hue, wheel saturation, and luminance controls. Preview uses native media playback, while export uses a separate deterministic offline rendering path with WebCodecs MP4/WebM composition, shared geometry, audio mixing, captions, overlays, effects, and a MediaRecorder fallback. Audio clips or complete timeline mixes can be exported as WAV or MP3 with trimming, speed, volume, fades, and spatial effects applied. The app is installable as a PWA with a cached shell and multilingual interface. The browser AI toolset includes multilingual voiceover with Chinese, mixed Chinese/English, English, German, Spanish, French, Italian, and Brazilian Portuguese voice options; Stable Audio-based local music generation with prompt translation and selectable durations; Whisper-based automatic captions; YOLOS and MODNet-based smart framing, caption avoidance, portrait matting, and background removal; MI-GAN watermark/object removal with timed regions; NanoVSR restoration; vocal separation; and JoyVASA plus LivePortrait digital-human generation. Large models are lazy-loaded, revision-pinned, and cached by the service worker. Supported workflows can run without uploading project media to an editing backend. Models are mirrored on Hugging Face and ModelScope with automatic source fallback and shared cache identities. For AI agents, the repository includes an editing skill for Codex, Claude Code, Copilot, and Gemini CLI. A versioned headless command runner can inspect projects and tracks, validate JSON edit plans, show predicted diffs, apply supported operations transactionally, perform dry runs, use idempotent operation IDs, import hashed media, and render a documented portable Visuals + Voiceover + Music subset to H.264/AAC MP4. The browser WebMCP integration exposes tools for structured inspection, reviewed timeline edits, asset and picture-in-picture insertion, rendered frame and audio sampling, local voiceover and transcription jobs, guarded undo, project saving, and export with progress, output metadata, and cancellation. The development setup requires Node.js 20+ and a modern Chromium browser, with WebGPU recommended for heavier AI workflows; the agent CLI documentation specifies Node.js 22.20.0 or later. The original source code is MIT licensed, while third-party models, weights, datasets, fonts, stock media, and other downloaded assets remain under their separate upstream licenses. The README also includes responsible-use requirements for deep-synthesis features, including authorized imagery and no illegal, misleading, or impersonating content.