rav1e

rav1e, (rav1e is a AV1 Encoder), the AV1 Encoder written in rust targeted for use-cases where libaom is very slow. There are other AV1 Open-Source Encoders like libaom from Google, SVT-AV1 from Intel and Netflix.

State-of-the-art decoders like dav1d are great for video playback, but best-in-class, free and open source software encoders are equally important to a healthy AV1 community. The AOMedia reference encoder was developed with the goal of creating the AV1 standard, not a production encoder. Thus, Mozilla and Xiph.Org are jointly developing a clean-room encoder named rav1e (the Rust AV1 Encoder) to increase encoding gains over the reference encoder and allow software encoding fast enough for real-time applications like WebRTC. Good encoders make heavy use of psychovisual models to allocate bits for what humans perceive as good visual quality (not PSNR). With rav1e we are applying the perceptual analysis expertise from our earlier Daala and Theora codec development efforts to add activity masking, better color balancing, improved rate control and perceptual distortion metrics like CDEF that bring new, improved quality to AV1 encoding.

rav1e is also investing considerable research to improve encoder speed, optimizing new techniques that appear for the first time in AV1. It’s not enough to rewrite existing code from the initial reference encoder in SIMD assembly and make it four times faster. Rav1e is developing ways to make AV1 encoding tools 1000x faster by finding new algorithms rather than simply optimizing existing code.

Rav1e is getting better all the time. Active development continues at a rapid pace, landing major new improvements weekly.