gamut-avif

gamut-avif is a pure-Rust, memory-safe AVIF encoder that wraps AV1 intra-frame bitstreams in an ISOBMFF/MIAF container.

This is the high-level crate most users want: give it pixels, get a complete .avif file. It is orchestration only — gamut-av1 does the AV1 coding and gamut-isobmff writes the container.

Goals

Part of the gamut workspace, this crate exists to provide AVIF encoding that is:

• Memory-safe on hostile input. #![forbid(unsafe_code)] end to end — the entire encode and container path is safe Rust, deleting the spatial/temporal memory-corruption bug class that has repeatedly bitten the C image codecs.
• Buildable anywhere cargo is. No C, no autotools/CMake, no nasm — just Rust, so it cross-compiles cleanly (wasm32, aarch64, musl) for serverless/edge image optimization.
• Encoder-first and size-first. The product is the encoder and the bytes it emits; the space/time tradeoff of each mode is documented as it lands.
• Clean-slate from the official specs. Implemented directly from the AV1 Bitstream & Decoding Process Specification and the AVIF / AV1-ISOBMFF bindings (see ../../references/), so it is auditable and forkable rather than a wrapper over libaom/libavif.
• Permissively licensed (MIT OR Apache-2.0), matching the royalty-free AV1/AVIF formats.

It builds on the workspace's shared primitives: gamut-color (pixel formats / CICP), gamut-dsp (transforms), gamut-bitstream (bit writer + AV1 symbol coder), gamut-av1 (the AV1 keyframe encoder), and gamut-isobmff (the container).

Usage

use gamut_avif::AvifEncoder;
use gamut_core::{Dimensions, EncodeImage, ImageRef, Rgb8};

let width = 64;
let height = 64;
let rgb: Vec<u8> = vec![0; width * height * 3]; // 8-bit interleaved RGB

let dims = Dimensions { width: width as u32, height: height as u32 };
let mut avif = Vec::new();
AvifEncoder::new()
    .encode_image(ImageRef::<Rgb8>::new(&rgb, dims).expect("dimensions"), &mut avif)
    .expect("encode");
std::fs::write("out.avif", &avif).unwrap();

AvifEncoder implements [gamut_core::EncodeImage<Rgb8>], so the input is a typed [gamut_core::ImageRef] and handing it an unsupported pixel layout is a compile error.

Status

Today (milestone M0) the encoder produces lossless still images: 8-bit RGB mapped to AV1 identity-matrix 4:4:4 (so the decoded image is bit-exact to the input), wrapped as a single av01 item. The space/time tradeoff is the obvious one — lossless output is exact but large; it makes no attempt to be compact yet, and correctness is the priority. Output is verified bit-exact against real decoders (libavif, dav1d), linked from vendored third_party/ submodules rather than system-installed binaries.

Everything beyond M0 — lossy intra (DCT + quantization), alpha, HDR/wide-gamut, 10/12-bit and 4:2:0/4:2:2, image sequences, and the rest of the AV1/AVIF surface — is tracked, row by row against the relevant specs, in STATUS.md.

License

Licensed under either of MIT or Apache-2.0 at your option.