Capped CRF is the best-of-both-worlds rate-control mode for streaming video. Plain crf targets a quality level and lets the bitrate fall wherever — sometimes very low on simple scenes, sometimes very high on busy ones. Plain VBR targets an average bitrate and lets quality vary. Capped CRF combines them: target a quality level but cap the maximum bitrate so the worst-case scenes don't blow your CDN budget or oversaturate a viewer's network connection.
The mechanics: the encoder behaves like CRF — picking the bitrate that hits the chosen quality on each scene — until the bitrate would exceed the cap. At that point the cap kicks in and the encoder behaves like CBR for those scenes, accepting a bit of quality loss in exchange for predictability. On easy content (talking heads, static UI), the cap never engages and you save bandwidth automatically. On hard content (snowstorm action, sports), the cap engages and quality matches what a CBR encode would have produced anyway.
For a streaming product, capped CRF has become the default rate-control mode for VOD adaptive bitrate ladders in 2026. It gives you the bandwidth savings of CRF on the 60–70 % of content that's easy to encode, the bitrate predictability of CBR on the 20–30 % that isn't, and a single setting works across an entire library. Typical configuration: pick a target CRF (22–25 for H.264, 25–28 for HEVC, 28–32 for AV1) and a cap of 1.5–2× the average bitrate you'd want for that rendition. Both x264 and x265 support it natively, FFmpeg exposes it via -crf and -maxrate flags, and Netflix's per-title encoding pipeline uses a sophisticated version of it.
