FFmpeg 9.0: WebP in and GIF out

FFmpeg 9.0 has been officially released, marking a major update to the widely used open-source multimedia framework. The new version introduces support for animated WebP files, enhances hardware acceleration capabilities, and adds features tailored for high dynamic range (HDR) and AI-driven video processing pipelines. FFmpeg, which provides tools and libraries for manipulating audio and video data, continues to serve as a foundational component in many desktop applications and server-side systems that automatically process uploaded media files. The key innovation in this release is the inclusion of support for animated WebP. FFmpeg now includes a dedicated decoder capable of converting compressed image data into individual frames, enabling further processing by the software. In conjunction with a new demuxer, the system can extract the file structure and timing information from animated WebP files, allowing users to convert them into GIFs or other formats. This enhancement expands FFmpeg’s versatility in handling modern web-compatible video formats. Hardware acceleration has also received significant improvements. The new version supports ProRes RAW decoding on Apple systems through the VideoToolbox framework. Additionally, FFmpeg 9.0 enables hardware-accelerated decoding of the Advanced Professional Video Codec (APV) using Vulkan. New GPU-based filters such as transpose_cuda, which allows rotation and mirroring of videos, and v360_vulkan, which facilitates conversion between different 360-degree projection types, have been added. These updates aim to improve performance and efficiency in video processing workflows. Support for dynamic HDR metadata via SMPTE-2094-50 has been enhanced, allowing FFmpeg to pass and process these scene- or frame-level parameters. This ensures compatible displays can adjust their output accordingly. A new bitstream filter for multi-layered Dolby Vision HEVC content enables the software to separate base video and additional Dolby Vision data for independent processing. The MP4 muxer now supports writing LCEVC tracks, expanding its utility in creating advanced video files. For AI-assisted video processing, FFmpeg 9.0 integrates support for ONNX Runtime as a backend for deep-learning models. This integration allows direct incorporation of neural network functions into the pipeline, including GPU execution providers. Filters within FFmpeg can apply neural networks to individual frames, supporting tasks such as upscaling, noise reduction, object recognition, and classification. This marks a step toward more intelligent and automated video editing capabilities. AMD's Advanced Media Framework (AMF) has also seen new extensions. The filter vf_vpp_amf handles HDR video conversion while considering color and HDR information. vf_frc_amf is designed for frame rate conversion, and vf_vqe_amf aims to enhance image quality. By leveraging AMF, FFmpeg can offload these functions directly onto supported AMD GPUs, improving performance and reducing CPU load during complex video operations. In terms of format and codec support, HE-AAC with 960-sample frames has been introduced, a variant crucial for the DAB+ radio standard. It specifies the 960-sample transform for MPEG-4 HE-AACv2, ensuring accurate decoding of such streams. Additionally, the Playdate handheld console benefits from a new encoder and muxer included in FFmpeg 9.0. Several outdated components and interfaces have been removed, including the CELT decoder and Ogg/CELT parsing, although Opus remains unaffected. Legacy NVENC options have also been phased out, with support for NVIDIA NVENC SDKs prior to version 11.1 ending with this release. Users should review the updated changelog on the project’s website for detailed information on all changes.