@video-editor/media

Media decode/encode abstraction for the video editor, currently implemented on top of mediabunny (WebCodecs).

Why this package exists

protocol and renderer must not depend on a concrete third-party media library. This package owns that dependency and exposes a small, stable contract; swapping the underlying library only ever touches this package.

Contract

• All timestamps on the public API are milliseconds (matching the video protocol). Conversion to the underlying seconds-based API happens inside.
• openMediaInput(blob | url) returns a MediaInputHandle:
  • meta() — duration/dimensions/audio info
  • canDecodeVideo() / canDecodeAudio() — WebCodecs support probe; callers must report an unsupported-media error when a required decoder is absent
  • drawFrame(ctx, timeMs) — decode one frame and draw it onto a canvas; frame lifecycle (close()) is handled internally
  • thumbnails(width, { startMs, endMs, stepMs })
  • decodeAudioSlice(startMs, endMs) — one AudioBuffer for WebAudio use
  • dispose()
• createMp4Encoder({ canvas, ... }) returns a streaming fMP4 encoder driven by an external render loop: addFrame(tMs, durMs) captures the canvas, setAudio(buffer) encodes a mixed-down track, stream yields the file bytes.
• renderCanvasToVideo({ canvas, durationMs, fps, renderFrame, sink }) drives an HTML or offscreen canvas with an exact media clock and writes MP4 or WebM bytes without buffering the whole file in memory.
• captureCanvasStream({ canvas, frameRate }) returns a real-time MediaStream. HTML canvases use captureStream(); offscreen canvases use VideoFrame with a browser video-track generator. Missing browser features are reported before capture starts.
• trimVideo({ source, startMs, endMs, sink }) trims the primary video and audio tracks into a streaming MP4 or WebM output. The result reports whether each track was copied or transcoded.