HDR video: PQ, HLG and the metadata problem
What HDR actually changes in the signal, why HDR content looks grey on SDR screens, and which of the four formats you need.
Updated
HDR is sold as “brighter”, which undersells it and explains why so many HDR workflows go wrong. What actually changes is the meaning of the numbers: an HDR signal maps its code values to real-world luminance in nits, where an SDR signal only ever described a relative scale.
That reinterpretation is why HDR content played as if it were SDR looks flat and washed out, and why the failure is so recognisable once you have seen it.
The transfer function is the change
Every video signal carries a transfer function: the curve relating stored values to light. SDR uses a gamma curve inherited from cathode-ray tubes, targeting a display around 100 nits, and it describes brightness relatively. 100 percent means “as bright as this screen goes”.
HDR replaces it with one of two curves:
PQ (Perceptual Quantizer, SMPTE ST 2084) is absolute. A given code value means a specific luminance, up to 10,000 nits, whatever the display. That precision is why PQ dominates streaming and cinema, and it is also why PQ content needs to know about the display it will land on, which is the whole metadata story below.
HLG (Hybrid Log-Gamma), developed by the BBC and NHK, is relative and backwards compatible by design. The bottom of its curve is ordinary gamma, so an HLG signal shown on an SDR screen looks approximately correct rather than broken. That property is worth a great deal in live broadcast, where one feed serves every kind of receiver, and it is why HLG owns broadcast while PQ owns streaming.
The other two ingredients
HDR is not only the curve. A complete HDR signal also needs:
A wide colour primary set, normally Rec. 2020, which describes a much larger set of colours than SDR’s Rec. 709. Get this wrong independently of the transfer function and you get oversaturated or undersaturated colour on top of whatever the curve is doing.
10-bit depth, at minimum. Stretching a signal across a 10,000-nit range with only 256 steps per channel would band catastrophically. HDR and 10-bit are inseparable in practice, which is why HEVC Main 10 and AV1 became the HDR codecs and 8-bit H.264 never could be.
The four format names
The formats people argue about differ mainly in metadata, not in the signal:
HDR10 is the baseline: PQ, Rec. 2020, 10-bit, plus static metadata describing the mastering display once for the whole title. Open, royalty-free, and supported by essentially every HDR device. If you ship one HDR format, this is it.
HDR10+ adds dynamic metadata, per-scene, so a dark scene and a bright scene can be tone-mapped differently. Royalty-free, backed by Samsung, with meaningful but partial support.
Dolby Vision also uses dynamic metadata, is licensed, and has the widest studio and streaming-service adoption of the dynamic formats. It also supports 12-bit.
HLG is the broadcast one described above, and carries no dynamic metadata by design.
For a service starting today, HDR10 plus SDR fallbacks covers the field; Dolby Vision is the one you add when the content and the licensing justify it.
Why it goes wrong
Tone mapping is where HDR meets reality. Almost no display reaches 10,000 nits; a good consumer TV manages 600 to 1500. So every HDR playback path tone-maps the signal down to what the panel can do. Dynamic metadata exists to make that mapping smarter per scene. When tone mapping is absent or naive, HDR content looks dim and muddy rather than spectacular.
Playing HDR as SDR is the classic bug. If the transfer function is not signalled, or a tool strips the metadata during a remux, the player applies an SDR gamma curve to a PQ signal. The result is the washed-out, grey-blacks look described in colour spaces and washed-out video, in an especially severe form.
Converting HDR to SDR properly is not a filter you apply casually. A naive
downconvert clips highlights or crushes the image. FFmpeg can do it with the
zscale and tonemap filters, and the parameters matter; every shortcut here
shows.
Browsers are the weakest link. HDR video on the web remains patchy and platform-dependent in a way that HDR on TVs and phones no longer is. If your delivery target is a browser, plan for an SDR version and treat HDR as an enhancement, not a baseline.