The FFmpeg project has released FFmpeg 9.0, the latest major update to the widely used open-source multimedia framework, adding animated WebP decoding, more GPU-accelerated processing options, expanded HDR handling, and support for several newer media workflows.
One highly anticipated addition is native support for animated WebP files via a new decoder and demuxer, which lets FFmpeg read animated WebP content directly. Additionally, FFmpeg 9.0 adds APV hardware acceleration using Vulkan and introduces VideoToolbox-accelerated ProRes RAW decoding on supported Apple hardware.
The release also introduces the v360_vulkan filter, a Vulkan-accelerated implementation for processing 360-degree and panoramic video. NVIDIA CUDA users gain the new transpose_cuda filter, allowing frame rotation and transposition to stay on the GPU instead of moving frames back to system memory for software processing.
AMD’s Advanced Media Framework support has expanded as well. The existing vf_vpp_amf color-conversion filter gains broader HDR capabilities, while a new vf_frc_amf filter provides hardware-assisted frame-rate conversion. FFmpeg 9.0 also includes support and passthrough for SMPTE 2094-50 dynamic metadata, improving handling of metadata-driven HDR workflows.
On the audio side, the release adds HE-AAC 960 decoding for DAB+ content. Unlike the more common 1024-sample AAC frame configuration, some digital radio services use 960-sample frames, and the new support improves compatibility with those broadcasts.
FFmpeg’s container and delivery capabilities expand with support for muxing LCEVC tracks into MP4 files. Low Complexity Enhancement Video Coding includes an enhancement layer alongside a base video stream, enabling compatible decoders to reconstruct a higher-quality version while retaining backward compatibility.
Another addition targets Panic’s Playdate handheld console. FFmpeg 9.0 includes a Playdate video encoder and a corresponding muxer, letting users and developers to create video files suitable for the device directly through FFmpeg.
The release further adds a bitstream filter for splitting multilayer Dolby Vision HEVC streams. Machine-learning-based media processing also improves with an updated ONNX Runtime DNN backend that supports GPU execution providers, enabling compatible FFmpeg filters and applications to run neural-network workloads on available GPU backends rather than relying exclusively on CPU execution.
Finally, FFmpeg 9.0 removes several obsolete or deprecated components. The old standalone CELT decoder and Ogg/CELT parser are gone, although this does not affect CELT audio carried as part of the modern Opus codec. Deprecated NVENC options and support for NVIDIA Video Codec SDK releases older than version 11.1 have also been removed.
For additional details, see the changelog.
FFmpeg 9.0 is available now as a source-code archive from the project’s official download page. Binary availability will rely on individual Linux distributions and third-party package providers, with rolling-release distributions likely to receive the update first.
