FFmpeg 8.0 'Huffman' Released, Biggest Update Ever, Including Transcription AI 'Whisper' and Official Support for Vulkan-Based Codecs

Version 8.0 'Huffman' of the multimedia framework FFmpeg was released on August 22, 2025. This version adds the Whisper filter, an open-source AI speech recognition model, enabling automatic speech recognition within FFmpeg itself, and enabling features such as live subtitle generation and transcription.

FFmpeg
https://ffmpeg.org/index.html#pr8.0

FFmpeg/Changelog at master · FFmpeg/FFmpeg · GitHub
https://github.com/FFmpeg/FFmpeg/blob/master/Changelog

This update adds many new features, including new native decoders for APV, ProRes RAW, RealVideo 6.0, and more. The Versatile Video Coding (VVC) decoder has also been improved with features like IBC and Palette Mode.

Hardware acceleration has also been significantly improved, with support for Vulkan VP9, VAAPI VVC, and OpenHarmony H.264/5 for decoding, and Vulkan AV1 and OpenHarmony H.264/5 for encoding.

Another major feature of this release is the addition of a new class of decoders and encoders based on a pure Vulkan implementation. These do not rely on specific hardware accelerators and work using GPU compute shaders on any Vulkan 1.3 implementation. Currently, this new class supports FFv1 (encode and decode) and ProRes RAW (decode only).

And as previously announced, a new filter, Whisper, has been added, which takes advantage of FFmpeg's powerful video and audio processing capabilities and enables more advanced processing, such as transcribing audio and video files directly from the command line.

FFmpeg will add 'transcription function using OpenAI's Whisper' - GIGAZINE

However, please note that Whisper support depends on the FFmpeg build configuration, so it may not necessarily be included in the package you are using. You can check whether the Whisper filter is included in your environment with 'ffmpeg -filters | grep whisper.' When building from source yourself, you will need to enable the option '--enable-whisper' and specify the correct path to the model.