HDR fundamentals for delivery: PQ, HLG and metadata

CINE-202 introduced high dynamic range and its standards, and CINE-310 showed how a colorist places a grade in HDR and trims it for SDR. Delivery adds a different question: what exactly is inside the HDR file, and how will a television, a tablet or a phone interpret it? This lesson looks at the numbers behind HDR delivery. It covers the PQ and HLG transfer functions, the difference between the container color space and the mastering color space, code values and banding, and the static and dynamic metadata that travel with an HDR master. These details decide whether the image the cinematographer approved in the grading suite is the image viewers see at home.

Display-referred signals and absolute brightness

In SDR, the signal is relative. A code value near the top of the range means "as bright as this display goes", whether that display reaches 100 nits or 400 nits. PQ HDR works differently. The Perceptual Quantizer, standardized as SMPTE ST 2084 and included in ITU-R BT.2100, assigns each code value an absolute luminance in candelas per square meter (nits), from 0 up to 10,000 nits. A PQ code value of about 0.508 of the full range always means about 100 nits, and about 0.752 always means about 1,000 nits. This is useful for mastering, because the colorist knows exactly what brightness each part of the image has. It also means that televisions which cannot reach the mastered brightness must map the signal down to what they can show, which is where metadata helps.

The PQ curve

PQ was designed to match human contrast sensitivity, so that each code step is roughly equally visible from deep shadows to bright highlights. Its inverse electro-optical transfer function turns luminance into a signal value between 0 and 1.

Here \( L \) is luminance in nits and the constants are \( m_1 = 0.1593 \), \( m_2 = 78.84 \), \( c_1 = 0.8359 \), \( c_2 = 18.85 \) and \( c_3 = 18.69 \). You will never need to calculate this by hand on a job, because grading software and scopes do it for you, but working through it once makes code values and nit levels easier to read.

Code values, bit depth and banding

A 10-bit signal has 1,024 code values per channel. In narrow range, which most professional video uses, only 64 to 940 carry picture, giving 877 usable steps. Because PQ spreads those steps across 0 to 10,000 nits, each step near 100 nits is about 1 % brighter than the one below, and each step near 1,000 nits is also about 1 % brighter, roughly 9 nits. That is close to the threshold at which people can see a step in a smooth gradient on a large, bright display. This is why HDR delivery requires 10-bit or better signals, why 12-bit is used internally in mastering and Dolby Vision, and why heavy compression or 8-bit intermediate files cause visible banding in skies, smoke and dark walls.

HLG: hybrid log-gamma

Hybrid log-gamma (HLG), developed by the BBC and NHK and also part of BT.2100, takes a different approach. It is scene-referred and relative: the signal describes scene light, and each display maps it to its own brightness. The lower part of the curve behaves like a conventional gamma curve, so an HLG signal looks reasonable on an SDR display without metadata. The upper part is logarithmic to hold highlights.

The constants are \( a = 0.17883 \), \( b = 0.28467 \) and \( c = 0.55991 \). At \( E = 1/12 \), both parts give \( E' = 0.5 \), so the curve is continuous.

HLG is popular in live broadcast and sports, where there is no time for a separate SDR grade and metadata is hard to manage. Premium streaming originals are usually mastered in PQ, but many services accept or create HLG versions for some uses, and some productions deliver HLG for broadcast partners.

Container gamut and mastering gamut

HDR files are usually labeled Rec. 2020 (BT.2020) primaries with a PQ transfer function. Rec. 2020 is a very wide gamut that most displays cannot fully reproduce. In practice, most reference displays used for mastering cover the P3 gamut with a D65 white point, so colorists grade within P3-D65 and the file is placed in a Rec. 2020 container. Some services ask for P3-D65 masters in a Rec. 2020 container explicitly and limit colors to the P3 triangle. Others accept true P3-D65 PQ masters as an intermediate and convert later.

• Container: the color space the file declares, usually Rec. 2020 primaries and PQ.
• Mastering gamut: the colors actually used, usually limited to P3-D65 by the reference display.
• Out-of-gamut colors: values outside the mastering gamut that nobody saw on the reference display. QC tools flag them.

The cinematographer does not usually set these labels, but should ask the colorist and finishing facility to confirm the container and mastering gamut early, so that LUTs and monitors on set match the final pipeline.

Static metadata: HDR10

HDR10 is the most common baseline HDR format. It uses PQ, Rec. 2020 primaries, 10-bit video and static metadata that describes the whole program.

• Mastering display color volume (SMPTE ST 2086): the primaries, white point and minimum and maximum luminance of the reference display used for mastering, for example P3-D65 with 0.0001 to 1,000 nits.
• MaxCLL (maximum content light level): the brightest single pixel anywhere in the program, in nits.
• MaxFALL (maximum frame-average light level): the highest average brightness of any single frame, in nits.

A television uses these values to decide how to tone-map the program to its own capability. Because the values are fixed for the whole title, a single very bright frame can change how the entire program is mapped. Lesson 5 shows how MaxCLL and MaxFALL are measured.

Dynamic metadata: HDR10+ and Dolby Vision

Dynamic metadata describes the picture shot by shot or frame by frame, so a television can tone-map each scene differently.

• Dolby Vision: a proprietary system that adds shot-by-shot metadata generated by analysis and adjusted by the colorist with trims for target displays. Many premium services deliver in Dolby Vision; Lesson 5 covers its metadata levels.
• HDR10+: an open dynamic-metadata system based on SMPTE ST 2094-40, supported by some televisions and services.
• HLG: has no metadata in its basic form; displays rely on the relative signal.

A single PQ master can carry Dolby Vision metadata, from which HDR10 and SDR versions can be derived. This is one reason many services ask for a Dolby Vision master with its metadata file even when they also distribute HDR10.

Putting the pieces together

A typical premium HDR master is a 4K (3,840 × 2,160) picture with PQ transfer, Rec. 2020 container primaries, a P3-D65 mastering gamut, a mastering display peak of 1,000 nits, 12-bit or 16-bit image files in the finishing facility and 10-bit or 12-bit compressed or lossless delivery files, plus Dolby Vision metadata, HDR10 static metadata and an SDR version. These values are common examples; always follow the current specification. The cinematographer's practical takeaway is simple: know the mastering peak and gamut, light and expose so that important detail sits comfortably inside them, and ask to see the image on a calibrated HDR monitor before signing off.