#Reading and Owning Samples PRO

await MediaSampleReader.create(asset, "video" | "audio", { decoded: true, trackIndex: 0 }) opens one track. decoded: true yields BGRA video or interleaved Float32 PCM. Set decoded: false to preserve compressed packets and original codec descriptions; decoding support depends on the system. This does not change existing AVAssetReader image/array outputs.

copyNextSampleBuffer() returns an owned MediaSample or null only at successful EOF. Cancellation and decoding errors reject. Only one read may be pending per reader. Samples expose mediaType, presentationTime, decodeTime, duration, sampleCount and disposed; time fields use MediaTime, preserving native precision. Invalid/indefinite duration is possible and must be checked. Retain handles only as long as necessary and call dispose() in finally after enqueue settles. Native queues retain independently.

On iOS 27, ScreenCapture.Sample.toMediaSample() creates an independent sample handle without converting through UIImage/JS arrays. Only complete video frames can be converted. Capture source permissions still apply, and capture timestamps must be anchored to the playback clock. Disposing the source sample does not dispose its converted handle.

A script may retain up to 128 samples / 64 MiB of sample payload. Renderer queues have separate limits. These are managed payload limits, not a bound on codec/system memory. Handles belong to one script and cannot be transferred to another executor.

Passthrough readers preserve zero-payload drain/empty-edit markers. isMarker identifies these; enqueue them in sequence too, and check timing.isNumeric before using their timestamps.

#index.tsx

Open the documentation’s example entry to run the demo, or view the complete index.tsx.