HEVC (H.265) — the codec everyone uses and nobody loves

HEVC, also known as H.265, is the codec the entire streaming industry uses while loudly complaining about it. ITU-T and MPEG finalized it in 2013. Apple shipped FairPlay-encrypted HEVC for iOS streaming in 2017. Smart TVs picked it up by 2018. By 2020, HEVC was the de facto standard for premium streaming, broadcast, and storage. The patent licensing situation is genuinely bad. The codec itself is excellent. Both things are true.

This page is the engineering reference: what HEVC is, why it won despite the licensing mess, and where it sits in a 2026 ladder.

#What HEVC is

HEVC is a block-based hybrid video codec — same fundamental shape as H.264, with each pipeline stage made more sophisticated:

• Coding tree units (CTUs) up to 64×64, with quad-tree partitioning down to 8×8. H.264's macroblocks were 16×16 with limited subdivision; HEVC's variable structure is much better at adapting to content edges.
• Wider intra prediction — 33 angular modes plus DC and planar (vs H.264's 9). Better in-frame prediction means smaller residuals to encode.
• Improved inter prediction — motion vectors with quarter-pel precision (same as H.264) but better candidate prediction (advanced motion vector prediction, AMVP). Less bitrate spent on motion vector coding itself.
• Sample adaptive offset (SAO) — a deblocking-adjacent filter applied after the standard deblocker. Reduces the visible blocking and ringing that low-bitrate H.264 is notorious for.
• Tiles and slices — explicit parallelism support. Encoders can split frames into independently-encodable tiles for multi-threaded encoding without quality loss.
• Higher bit depths and chroma formats — 10-bit and 12-bit profiles standardized. 4:2:0, 4:2:2, 4:4:4 supported. The HEVC Main 10 profile is what HDR signaling rides on.

Compression-efficiency benchmarks against H.264:

• HEVC Main vs H.264 High: 40-50% bitrate reduction at equivalent visual quality.
• HEVC Main 10 vs H.264 High 10: similar, sometimes more, on HDR content.

The "50% improvement at half the bitrate" line is the marketing summary. It's roughly true on average content; some content sees less, some more.

#x265 in production

x265 is the open-source HEVC encoder maintained by MulticoreWare. It's the de facto reference for "real-world HEVC encoding" the way x264 is the de facto reference for H.264. There's a closed-source DVD/Blu-ray-grade encoder (Beamr) and several silicon-vendor encoders, but x265 is the one most production pipelines run.

The presets matter:

• ultrafast to veryfast — live encoding territory. Quality drops noticeably.
• fast — live with quality budget. Acceptable for most live streaming.
• medium — VOD baseline. Roughly comparable to libx264 medium in encoding time, with HEVC's compression advantage.
• slow — better quality for premium VOD. ~2-3× slower than medium.
• slower, veryslow, placebo — research-grade. Not realistic for production at any scale.

The CRF range for HEVC is different from H.264 — CRF 22 in x265 is roughly equivalent to CRF 18 in x264. Teams transitioning often serve over-encoded HEVC for the first few months because they didn't recalibrate the CRF.

Production tuning specifics worth knowing:

• Tune options — --tune zerolatency for live, --tune fastdecode to reduce decoder complexity (worth ~10% bitrate at the same quality, useful for low-power decoders), --tune grain for film-grain content.
• Look-ahead and B-frames — defaults are reasonable. Increasing --bframes past 4 rarely helps; increasing --rc-lookahead past 40 is diminishing returns.
• AQ modes — --aq-mode 3 (autovariance) is the default and usually correct. --aq-mode 4 (autovariance with bias to dark) helps on cinema-grade content.

#Hardware encode/decode

HEVC is the most-deployed hardware-accelerated codec on the planet. Every meaningful device shipped since ~2018 has hardware HEVC decode:

• Mobile — every iPhone since iPhone 6 (2014, hardware decode) and iPhone 7 (2016, encode). Every Android flagship since 2016. Practical install base: 100% for new traffic, ~95% for total install base.
• Desktop — every Intel CPU since Skylake (2015) has Quick Sync HEVC. Every NVIDIA GPU since Maxwell 2 (2014). Every AMD GPU since Carrizo (2015).
• TVs — every smart TV since 2017+ supports HEVC decode in hardware. Broadcast set-top boxes have shipped HEVC for longer.
• Browsers — Safari (macOS + iOS) supports HEVC natively. Chromium added HEVC decode in 2022 (gated by hardware). Firefox added it in 2024.

Hardware encode quality is generally 1-2 quality tiers behind x265 software at the same bitrate. NVENC HEVC quality has improved meaningfully (RTX 30+ is much better than earlier generations), but software still wins where wall time isn't the constraint. For live, hardware HEVC is the right answer; for VOD, software x265 is usually preferred unless throughput matters more than quality.

#The patent licensing situation

This is the actually-bad part. HEVC ships with three patent pools plus uncategorized patent holders, and the licensing math is genuinely difficult:

• MPEG LA HEVC license — the oldest pool, formed in 2014. Per-unit royalty for hardware decoders, per-stream consideration for streaming services. Many incumbent patent holders.
• HEVC Advance / Access Advance — formed in 2015 because some patent holders weren't satisfied with MPEG LA's terms. Higher rates, different licensee mix.
• Velos Media — formed in 2017 as a third pool with a different fee structure.
• Unaffiliated patent holders — Samsung, Apple, and several others have at times asserted patents outside the pools.

The result: a streaming service can't just sign one license and be done. Legal teams have to negotiate with multiple parties, and the per-stream costs at scale are nontrivial. This is exactly the situation that motivated AOMedia to form and produce AV1 as a royalty-free alternative.

In practice, the major streaming services (Netflix, Apple, Amazon, etc.) have settled licensing agreements with all the relevant parties. Smaller streamers and self-hosted deployments often skip the formal licensing on the gamble that the pools won't pursue them — a defensible position for non-commercial use, much riskier at commercial scale. If you're building a B2B product that uses HEVC, talk to a media-IP lawyer; if you're streaming consumer-facing at scale, you're probably already paying.

#When HEVC is the right answer in 2026

For most production streaming workflows, HEVC is still the default codec. Specifically:

• Apple ecosystem — Safari + iOS support HEVC natively but not AV1 (Safari iOS still doesn't ship AV1 decode in 2026). FairPlay DRM works with HEVC. Any pipeline that has to serve iOS users needs an HEVC ladder.
• Smart TV audience — the long tail of 2018-2021 smart TVs decode HEVC but not AV1. Still a meaningful slice of premium streaming traffic.
• Live streaming — hardware HEVC encode is mature and works at low latency. Live AV1 is possible but ecosystem-thin. Hardware HEVC is the live default.
• Broadcast contribution and distribution — DVB and ATSC 3.0 standardized on HEVC. Broadcast workflows are HEVC.
• Storage-bound archive — when you're transcoding once and storing indefinitely, HEVC's compression vs H.264 is real money. AV1 is better here, but the encoder maturity gap matters less for archive than for live.

#When to skip HEVC

• AV1-only premium tiers — if your audience is overwhelmingly Chromium browsers and recent Android/iOS Pro models, you can serve AV1 to them and reduce your HEVC burn.
• H.264-only floor — if you're streaming to developing-market devices or the very long tail of pre-2017 hardware, H.264 is still the answer. Don't add HEVC to a ladder that's reaching downlevel devices.
• Internal-only video — if you're doing internal video for a company intranet where you control the player, license cost dominates. AV1 (royalty-free, software decode acceptable) is often the answer.

#The HEVC ladder in 2026

The pragmatic 2026 ladder for premium consumer streaming:

1. AV1 top tiers — 4K, 1440p, 1080p where audience capability allows. Served to AV1-capable clients.
2. HEVC mid tiers — 1080p, 720p. The fallback for Apple/Safari, older Android, mid-range smart TVs. The bulk of streaming bandwidth.
3. H.264 floor — 720p down. The catch-all for downlevel devices.

HEVC is not going away. It's the middle tier of every consumer streaming ladder for the foreseeable future, and the "we ship to Apple" tier permanently or until Safari ships AV1. The licensing situation is unpleasant but workable; the codec itself is excellent.

#A note on HEVC vs AV1 vs VVC

The 2026 codec hierarchy looks roughly like this. AV1 is 30% better compression than HEVC, royalty-free, but lags HEVC on hardware install base and Apple ecosystem support. VVC is 30-50% better than HEVC, three-pool licensing, and effectively zero consumer hardware decode. HEVC is the middle of the pyramid: codec quality is good, hardware support is universal, and the licensing pain is annoying but priced into industry economics. For consumer streaming in 2026, the ladder is AV1 top + HEVC middle + H.264 floor. HEVC is the codec that makes the ladder work because it covers the audience neither AV1 nor H.264 reaches well — Apple devices, older smart TVs, the broad middle of hardware capability.

#What MpegFlow does with HEVC

MpegFlow's DAG runtime runs x265 on FfmpegExecutor workers with per-tier preset configuration in workflow YAML. Hardware HEVC encode (NVENC on the GPU pool, Quick Sync where the worker hardware supports it) is configurable for live workflows where wall time matters more than peak quality. The partitioner persists each rendition stage to job_stages with explicit dependency tracking; per-stage retry handles transient failures; sibling cancellation propagates fatal failures across rendition stages.

Default ladder templates include HEVC mid tiers because that's the production reality for any pipeline serving Apple devices or older smart TVs. The KEDA-driven autoscaler sizes the GPU and CPU worker pools independently to the queued workload.

For HEVC HDR (HDR10, HDR10+, Dolby Vision), the metadata signaling required by playback ecosystems flows through the encoder via the appropriate FFmpeg flags; the pipeline preserves the SEI metadata end-to-end through timestamp-preservation discipline at every stage boundary. Note that Dolby Vision Profile 5/8 mezzanine workflows requiring metadata signaling beyond what x265 emits natively are operator-side work today; the worker image runs the standard FFmpeg/x265 path.

If you're sizing an HEVC deployment and want a sounding board on encoder choice, ladder design, or the licensing question, we run that conversation regularly.