Skip to content

v6.1.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 27 Jun 04:34
· 193 commits to main since this release
v6.1.0
a5f5d67

Added

  • Typed context / graph field bags and a configure hook on Encoder, Decoder, FilterAPI, FilterComplexAPI, Muxer and Demuxer. context (encoder/decoder/muxer) and graph (filters) are typed bags of the writable fields on the underlying native object (e.g. context: { bitRate: '5M', gopSize: 60, colorPrimaries: AVCOL_PRI_BT2020 }); rate-control fields still accept '5M'-style strings. configure(ctx | graph | fmt) is a callback for what the bag cannot express — flags via setFlags, methods, or setter-only forms such as a 'hvc1' FourCC string for -tag:v hvc1. Both run right before the object is opened/finalized (avcodec_open2, avfilter_graph_config, avformat_write_header), the bag first; for the muxer after the source metadata/disposition are copied, so values set there take precedence. The Demuxer exposes only configure(fmt), called after avformat_open_input / avformat_find_stream_info.

    const encoder = await Encoder.create(FF_ENCODER_LIBX265, {
  decoder,
  context: { bitRate: '5M', gopSize: 60 },        // typed, declarative field bag
  configure: (ctx) => { ctx.codecTag = 'hvc1'; }, // imperative escape (-tag:v hvc1)
});

  • HardwareContext.auto() accepts a prefer option to prioritise specific hardware type(s) over the platform default order, e.g. auto({ prefer: AV_HWDEVICE_TYPE_VAAPI }) to pick VAAPI ahead of QSV on Intel. Preferred types are tested first (and kept only if available), then the usual platform fallback applies.

  • Bitstream filter extradata is now reflected in the output parameters. Filters that adapt container framing (aac_adtstoasc, extract_extradata, dump_extra, *_mp4toannexb) emit the resulting global headers as packet side data rather than on par_out. BitStreamFilterAPI.outputCodecParameters now folds that side data in, so a stream copy that relies on it (including Muxer's bsf option) writes a header with the correct codec configuration.

Changed

  • Synced FFmpeg with the latest master (~380 upstream commits) — numerous bug fixes, stability and performance improvements. Highlights relevant to node-av users:
    • Security hardening — ~29 overflow / bounds fixes across decoders, demuxers and filters
    • swscale PAL8 — palette video handled by the new ops scaler engine
    • AMF — zero-copy host memory mapping, HDR/colour metadata propagation, high_quality encoder preset
    • NVENC — AV1 hierarchical B-frame reference mode
    • D3D11VA — native P016 (10/16-bit YUV 4:2:0) hardware decoding on Windows
    • ONNX Runtime DNN backend — GPU inference for ML-driven filters (super-resolution, denoise, …)
    • Dolby Vision — RPU detection across all NALs (better DoVi interop)
    • Fragmented-MP4 / DASH / HLS correctness fixes
    • Faster decode/encode on x86 (H.264 AVX2 intra-prediction, me_cmp SSSE3), AArch64 NEON and RISC-V RVV
  • Regenerated constants, encoders, and decoders from updated FFmpeg headers

Deprecated

  • Raw codec/graph field options moved to the context / graph bag. Encoder's bitrate / minRate / maxRate / bufSize (→ context: { bitRate, rcMinRate, rcMaxRate, rcBufferSize }, still accepting '5M'-style strings), gopSize, maxBFrames, threadCount, threadType; Decoder's threadCount / threadType; and FilterAPI / FilterComplexAPI's threads (→ graph: { nbThreads }), scaleSwsOpts, audioResampleOpts. They still work — set them via the bag instead.

Fixed

  • Rotation / HDR side data was lost when transcoding. The muxer copied a stream's coded_side_data (display matrix / rotation, HDR mastering display & content light level) only on stream copy, not when re-encoding — so a portrait phone video would play sideways and HDR metadata would vanish after a transcode. The encoder output path now copies it too, matching the stream-copy behaviour.
  • Demuxer dropped packets under backpressure in multi-stream pipelines (#263). When a single demuxer fed two consumers draining at different rates (e.g. a slow video chain alongside a fast audio chain), the demux thread discarded packets for whichever stream's queue was momentarily full instead of waiting for it to drain. This lost video reference frames and corrupted decoding (missing reference picture, mmco: unref short failure, co located POCs unavailable). The read loop now applies per-stream backpressure and is paced by the slowest consumer, so no packet is dropped.
  • worker_threads safety. Native class constructor references are now thread_local, so node-av objects can be created and used across multiple worker threads without crashing.
  • Cross-compiled Windows installs selected the wrong toolchain (#260). A cross install (npm install --os=win32 --cpu=x64) incorrectly picked the MinGW FFmpeg build; it now defaults to the standard MSVC build (pass --libc=mingw to override). See the new Cross-Platform Packaging section in the README.
  • Scaler failed on QSV hardware frames (Scaler: filter produced no frame). The Intel QSV scaler (vpp_qsv) buffers internally and does not emit a frame synchronously after a single input, so the scaler's process-one-receive-one path got nothing back. It now flushes the graph to drain the pending frame and rebuilds it for the next call. Zero-latency hardware scalers (scale_vaapi, scale_vt) are unaffected.

📋 All Changes since v6.0.0

📦 Packages

The following platform packages have been updated to v6.1.0:

  • @seydx/node-av-darwin-arm64@6.1.0
  • @seydx/node-av-darwin-x64@6.1.0
  • @seydx/node-av-linux-arm64@6.1.0
  • @seydx/node-av-linux-x64@6.1.0
  • @seydx/node-av-win32-arm64-mingw@6.1.0
  • @seydx/node-av-win32-x64-mingw@6.1.0
  • @seydx/node-av-win32-arm64-msvc@6.1.0
  • @seydx/node-av-win32-x64-msvc@6.1.0

Contributors

seydx
Assets 18
Loading