{"id":1044,"date":"2026-05-15T09:49:25","date_gmt":"2026-05-15T02:49:25","guid":{"rendered":"https:\/\/liveapi.com\/blog\/av1-encoding\/"},"modified":"2026-05-15T09:50:09","modified_gmt":"2026-05-15T02:50:09","slug":"av1-encoding","status":"publish","type":"post","link":"https:\/\/liveapi.com\/blog\/av1-encoding\/","title":{"rendered":"AV1 Encoding: How It Works, Encoders, and How to Use It"},"content":{"rendered":"Reading Time: <\/span> 14<\/span> minutes<\/span><\/span>

Netflix reported in late 2025 that around 30% of its streams now play back in AV1, and YouTube, Meta, and Twitch are pushing similar shifts. The reason is simple: AV1 cuts file sizes by 30\u201350% versus older codecs at matching visual quality, which translates directly into lower CDN bills and better playback on slow networks. The catch is that AV1 encoding is computationally expensive, the tooling is fragmented across three software encoders and three hardware vendors, and the decoder support story is still uneven on older devices.<\/p>\n

This guide walks through what AV1 encoding is, how the codec actually compresses video, the differences between libaom, SVT-AV1, rav1e, and the hardware encoders from Nvidia, Intel, and AMD, and the FFmpeg commands you need to start producing AV1 files today. It also covers the trade-offs between hardware and software encoding, the parameters that matter most, and how to pick an encoder for VOD, live, or real-time use cases.<\/p>\n

What Is AV1 Encoding?<\/h2>\n

AV1 encoding<\/strong> is the process of compressing raw video frames into the AV1 (AOMedia Video 1) bitstream format using an AV1 encoder. AV1 is an open, royalty-free video codec<\/a> developed by the Alliance for Open Media \u2014 a consortium that includes Google, Netflix, Amazon, Meta, Microsoft, Apple, Intel, Nvidia, AMD, and others \u2014 as the successor to VP9. The first AV1 bitstream specification shipped in March 2018, and Netflix became the first major service to deploy it in February 2020.<\/p>\n

What makes AV1 encoding distinct from older codecs is its blend of three properties: aggressive compression efficiency (30\u201350% smaller files than H.264 at similar quality), royalty-free licensing with no per-stream patent fees, and explicit design for internet video rather than broadcast. The trade-off is computational cost \u2014 early reference encoders ran 100\u2013600\u00d7 slower than H.264, though modern SVT-AV1 and hardware encoders have closed most of that gap.<\/p>\n

The output of AV1 encoding is an AV1-coded video bitstream wrapped in a container like MP4 (ISOBMFF), WebM, or Matroska. That bitstream can be decoded by any AV1-compatible player, browser, or hardware decoder.<\/p>\n

How Does AV1 Encoding Work?<\/h2>\n

AV1 follows the same block-based hybrid coding pipeline used by H.264, H.265, and VP9, but adds dozens of new prediction modes, partition shapes, and filter tools. At a high level, the encoder takes each frame, splits it into blocks, predicts each block from already-coded data, transforms and quantizes the prediction error, and packs the result into a compressed bitstream with an arithmetic coder.<\/p>\n

Here is what happens inside an AV1 encoder, step by step.<\/p>\n

1. Frame Partitioning<\/h3>\n

The encoder breaks each frame into 128\u00d7128 or 64\u00d764 pixel superblocks, then recursively splits them down to blocks as small as 4\u00d74 pixels. AV1 supports ten partition patterns \u2014 including the square splits used in VP9 plus T-shaped, horizontal, vertical, and 4:1\/1:4 rectangular splits. More partition shapes mean the encoder can match block boundaries to object edges more precisely, which reduces the residual that needs to be coded.<\/p>\n

2. Intra and Inter Prediction<\/h3>\n

For each block, the encoder picks the cheapest prediction mode. Intra prediction<\/strong> uses neighboring pixels in the same frame; AV1 supports 56 directional intra modes plus smooth, paeth, and chroma-from-luma predictors. Inter prediction<\/strong> uses pixels from previously coded frames; AV1 can pull from up to seven reference frames, supports warped and global motion, and adds overlapped block motion compensation (OBMC) for smoother boundaries.<\/p>\n

3. Transform Coding<\/h3>\n

The residual \u2014 the difference between the prediction and the actual pixels \u2014 gets transformed into the frequency domain. AV1 supports square and rectangular DCT, asymmetric DST (ADST), flipped ADST, and identity transforms. The encoder can pick separate horizontal and vertical transforms per block, which helps compress directional patterns like fabric, hair, or text edges.<\/p>\n

4. Quantization<\/h3>\n

The transform coefficients are quantized to throw away visually unimportant detail. AV1 uses a quantization step that varies per frame, per superblock, and per color plane, and includes optimized quantization matrices tuned for human perception. This is the main lever that trades file size against quality.<\/p>\n

5. Loop Filters<\/h3>\n

After reconstruction, AV1 runs three in-loop filters to clean up block boundaries and ringing artifacts: a standard deblocking filter, the Constrained Directional Enhancement Filter (CDEF), and a Wiener-based loop restoration filter. CDEF in particular helps preserve edge detail at low bitrates<\/a>.<\/p>\n

6. Entropy Coding<\/h3>\n

The final step packs everything \u2014 prediction modes, motion vectors, quantized coefficients \u2014 into a compressed bitstream using a non-binary arithmetic coder borrowed from the Daala codec. This coder is more hardware-friendly than VP9’s binary entropy coder and avoids the patent landmines around CABAC.<\/p>\n

7. Optional Film Grain Synthesis<\/h3>\n

AV1 has a unique trick for grainy or noisy content: it can strip film grain from the source, code the clean video efficiently, and resynthesize the grain at the decoder using a small set of parameters. This keeps the cinematic look without spending bits on coding random noise.<\/p>\n

AV1 Encoding vs H.264, H.265, and VP9<\/h2>\n

AV1 is not the only modern codec, and the choice between it and its competitors depends on your tolerance for encoding cost, decoder reach, and licensing complexity. Here is how the four main options compare for streaming video work.<\/p>\n\n\n\n\n\n\n\n\n\n\n
Attribute<\/th>\nAV1<\/th>\nH.265 \/ HEVC<\/th>\nH.264 \/ AVC<\/th>\nVP9<\/th>\n<\/tr>\n<\/thead>\n
Compression vs H.264<\/td>\n40\u201350% smaller<\/td>\n30\u201340% smaller<\/td>\nBaseline<\/td>\n25\u201335% smaller<\/td>\n<\/tr>\n
Licensing<\/td>\nRoyalty-free<\/td>\nPatent pools, per-unit fees<\/td>\nPatent pools, mostly resolved<\/td>\nRoyalty-free<\/td>\n<\/tr>\n
Browser support<\/td>\nChrome 70+, Firefox 67+, Edge 121+, Safari 17 (M3+\/A17+ only)<\/td>\nSafari, Edge; spotty in Chrome\/Firefox<\/td>\nUniversal<\/td>\nChrome, Firefox, Edge, Android<\/td>\n<\/tr>\n
Hardware decoder share (2025)<\/td>\n~30\u201340% of global devices<\/td>\n~70%<\/td>\nEffectively 100%<\/td>\n~80%<\/td>\n<\/tr>\n
Software encoder speed<\/td>\nSlow (real-time only on fast presets)<\/td>\nModerate<\/td>\nFast<\/td>\nModerate<\/td>\n<\/tr>\n
Best use<\/td>\nHigh-volume streaming, large libraries<\/td>\nApple ecosystems, 4K HDR<\/td>\nUniversal compatibility, fallback<\/td>\nExisting YouTube\/WebM pipelines<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

For a deeper breakdown of how AV1 stacks up against individual codecs, see AV1 vs H.264<\/a>, AV1 vs H.265<\/a>, and VP9 codec<\/a>.<\/p>\n

The practical takeaway: AV1 wins on file size and licensing, H.264 wins on universal playback, H.265 wins where Apple devices dominate, and VP9 is a stopgap if you already have a VP9 pipeline. Most production setups ship AV1 plus an H.264 fallback so older devices keep working.<\/p>\n

Types of AV1 Encoders<\/h2>\n

There are four meaningful AV1 encoder families today. Three are software encoders that run on the CPU and one is a category of hardware encoders built into modern GPUs and SoCs. Each has a different sweet spot.<\/p>\n

libaom-av1 (Reference Encoder)<\/h3>\n

libaom is the reference AV1 encoder published by AOMedia. It implements the full AV1 feature set and is the slowest of the bunch \u2014 single-threaded performance is often under 1 frame per second on a single CPU core at default settings. libaom is the gold standard for quality benchmarks and academic comparisons, but most production teams pick SVT-AV1 instead. In FFmpeg it is invoked as -c:v libaom-av1<\/code>.<\/p>\n

SVT-AV1 (Production Encoder)<\/h3>\n

SVT-AV1 (Scalable Video Technology for AV1) is the encoder that turned AV1 into a viable production tool. Originally built by Intel in collaboration with Netflix, it was adopted as AOMedia’s official next-generation encoder in 2020. SVT-AV1 offers 2\u20135\u00d7 the encoding speed of libaom at similar quality, scales across many CPU cores, and ships in FFmpeg as -c:v libsvtav1<\/code>. This is the right starting point for most teams.<\/p>\n

rav1e (Rust Encoder)<\/h3>\n

rav1e is a Rust-based AV1 encoder from the Xiph.Org Foundation and Mozilla. It prioritizes safety, multi-threading, and easy embedding in Rust applications. rav1e is slower than SVT-AV1 in most benchmarks but is the cleanest option if you need an encoder library inside a Rust codebase. FFmpeg invokes it as -c:v librav1e<\/code>.<\/p>\n

Hardware AV1 Encoders<\/h3>\n

Hardware AV1 encoders are fixed-function ASIC blocks built into modern GPUs and SoCs. The main implementations are:<\/p>\n