What just happened in codecs

jxl-rs v0.7.1 reduces binary size by about 100 kB by moving static data behind LazyLock into a Box. For JPEG XL tooling, smaller binaries make integration and distribution easier, especially where footprint and startup matter.

Mediabunny v1.55.7 ships a targeted fix for sample iterators not terminating when exceptions were thrown. That kind of iterator edge case can quietly stall long-running media jobs, so these patch releases are worth taking quickly.

Mediabunny v1.55.5 is out as a patch release. Keeping close to these point updates is practical if you depend on browser media pipelines, because they are where regression fixes and compatibility polish usually land first.

NVIDIA’s frame-rate up-conversion (FRUC) work being merged into FFmpeg’s Vulkan stack is notable because it pushes interpolation workflows closer to mainline tooling. When this kind of vendor capability lands upstream, it is much easier to test and compare in ordinary FFmpeg-based pipelines.

Intel’s ANV Vulkan driver picked up HuC support for Vulkan Video H.265 decoding, closing one of the practical gaps in Linux-side hardware decode. For media tooling that leans on Vulkan paths, this is the kind of plumbing work that turns feature checkboxes into something that actually runs on user machines.

A new FFmpeg issue reports a division-by-zero crash found by fuzzing. Bugs in decoder edge paths are exactly what media projects need surfaced early, because they harden the code paths that touch untrusted files in the real world.

Mozilla opened an intent-to-ship to enable JPEG XL by default in Firefox 157 across platforms. That moves JXL from an opt-in Labs toggle toward baseline browser behavior, which matters for anyone shipping high-density images where JXL’s quality-per-byte and decode speed are the point.

Kempf has spent two decades shipping VLC into a legal environment that wanted him to stop, so when he talks about software patents he is not speaking in the abstract. The interview covers why he sees AV1 as the first codec where the royalty-free promise is credible rather than aspirational, and where the remaining patent risk actually sits. Worth reading alongside the AV2 progress if you are trying to understand why the open-codec pipeline that feeds this whole ecosystem exists at all.

The Rust decoder sitting behind Chrome’s JPEG XL flag can now use more than one core, after a run of commits reworking the render pipeline to be thread-safe. Decode speed is one of the arguments a format has to win before a browser turns it on by default, so this is the unglamorous work that matters more than any announcement. The same release ports libjxl’s blue noise dithering for 8-bit output, which is what keeps gradients from banding on ordinary displays.

Vulkan encoding is how FFmpeg reaches your GPU’s encode block without a vendor-specific path, and its H.265 side had been trailing its own H.264 for no good reason. Tuning the session flags and CU sizes closed it: at 1080p the commit reports 289 to 358 fps against H.264’s 288 to 359, which is parity within noise. Merged to master rather than released, so it lands for most people in FFmpeg 9.1.

DTS is the audio you get off a Blu-ray, and it is not a codec browsers have ever cared about, which makes reading and writing it from JavaScript a genuinely odd capability to acquire. Mediabunny now handles it in ISOBMFF, Matroska and MPEG-TS, with a separate @mediabunny/dts package supplying a decoder and an encoder polyfill since no browser will do it for you. Useful if you have ever tried to get a disc rip into something a browser will play and found the audio, not the video, was the wall.

One addition, and a useful one: computeFrameRateMetrics() reports what a video track’s frame timing actually looks like rather than what its header claims. Declared frame rates lie routinely, through variable-rate phone capture, screen recordings that drop frames when nothing moves, and remuxes that inherit a number from somewhere else. Anything that has to make a decision from frame rate, ours included, has until now been trusting a field instead of measuring the packets.

Twenty-eight commits behind 9.0, and almost none of them are features: bounds checks in the bwdif filter, overflow rejection in the MPEG and HEVC muxers, zero-channel tracks refused, error recovery in the Dolby E parser. Point releases are where this work lands, and it matters most if you run FFmpeg over files you did not create, which is every upload pipeline. Nothing here changes a command you already run; it changes what happens when someone feeds you a malformed one.

The library behind every lab on this site, and this release lands a fix we have a direct stake in: encodability checks no longer throw on Safari in some cases. That check is what decides whether a codec appears greyed out or usable in the quantizer lab, the transcode lab and the encoder benchmark, so a throw there is the difference between a working page and a broken one for Safari visitors. Elsewhere: HLS single-file mode now writes segment files that are independently playable, and several fMP4 fragment bugs are gone for anyone writing sparse multi-track output.

Memory safety was Chrome’s stated condition for readmitting JPEG XL, and this pure-Rust decoder is the answer: vendored into Chromium in December 2025 and shipping behind Chrome’s flag since 145. Daily commits and a public issue tracker make this repo the closest thing to a live feed on JPEG XL’s browser future.