Skip to content

CONCEPT Cited by 1 source

HDR10+

Definition

HDR10+ is an open, royalty-free high-dynamic-range video format that extends the static-metadata HDR10 baseline with dynamic per-scene metadata describing peak luminance and colour volume characteristics, enabling playback devices to tone-map each scene independently to their own display capabilities. The practical difference is scene-adaptive vs. title-uniform tone mapping.

Why dynamic metadata matters

HDR content is mastered for a reference display — typically 1000–4000 nits of peak brightness. Most consumer displays peak much lower (300–1500 nits), so every HDR render on a consumer device is a tone-mapping operation: squeeze the mastering range into the display range without clipping or dulling the image.

HDR10 carries metadata once per title: a single MaxCLL (maximum content light level) + MaxFALL (maximum frame-average light level) for the whole movie. The device applies one tone-mapping curve to every scene — conservative enough to protect the brightest shot from clipping, which necessarily darkens every other scene.

HDR10+ carries metadata per scene (or more finely, per frame-group): the peak luminance of this scene, the distribution of bright pixels in this scene, the target tone-mapping curve for this scene. The device applies a different curve to every scene, so dark scenes stay dark, bright scenes don't clip, and mid-tone scenes retain detail.

The Netflix flashlight example

Netflix's 2025-12-05 AV1@30% post gives a concrete worked example to motivate the HDR10 vs HDR10+ choice (Source: sources/2025-12-05-netflix-av1-now-powering-30-of-netflix-streaming): a scene with a flashlight beam cutting across the frame.

  • HDR10. Device tone-mapping is titled to avoid clipping across the whole movie, which gets applied to the flashlight scene unchanged. The region under the flashlight beam over-exposes — "over-exposure of the region under the flashlight".
  • HDR10+. Per-scene metadata tells the device this scene has a very high peak luminance in a small region (the beam) — the device applies a more aggressive tone map just for this scene, so the flashlight beam's detail is preserved — "preservation of the flashlight detail".

Netflix: "We chose HDR10+ as the HDR format for its use of dynamic metadata, which enabled us to adapt the tone mapping per device in a scene-dependent manner."

Why HDR10+ specifically vs Dolby Vision

Dolby Vision is the other dynamic-metadata HDR format; it also carries per-scene or per-frame metadata and delivers comparable tone-mapping quality. The post does not explain the HDR10+-over-Dolby-Vision choice, but HDR10+ is:

  • Royalty-free and open, governed by the HDR10+ Technologies LLC consortium (Samsung + Panasonic + 20th Century).
  • Deployable in the bitstream without a per-stream licensing fee.
  • Consistent with Netflix's co-founding role in AOMedia and its preference for open / royalty-free media primitives.

Dolby Vision requires licensing fees and is a commercial competitor to HDR10+. Netflix ships both in practice on non-AV1 codecs, but chose HDR10+ for the AV1-HDR combination.

AV1 + HDR10+ at Netflix

  • Launched 2025-03-XX.
  • 85% of HDR catalogue (by view-hours) covered as of the 2025-12-05 post snapshot.
  • Netflix target: 100% within a couple of months — effectively complete by early 2026.
  • Motivation: HDR was already a premium Netflix feature since 2016; combining HDR with AV1's better compression
  • FGS-preserved grain gives "images with greater detail, more vibrant colors, and an overall heightened sense of immersion".

Relationship to reference metrics

Reference visual-quality metrics (VMAF / PSNR / SSIM) were designed for SDR content and are calibrated against SDR viewing conditions. HDR content evaluation is a separate methodology; VMAF comparisons quoted in the same Netflix post (AV1 +4.3 vs AVC, +0.9 vs HEVC) are computed across the whole session population, which is a mix of SDR and HDR sessions, but the HDR10+ adaptation story is orthogonal to VMAF — it's about per-device per-scene tone-mapping rather than encoder-quality comparison.

Seen in

Last updated · 319 distilled / 1,201 read