1 The idea
Scene values are relative; screens are measured in nits (cd/m²). An SDR screen tops out at about 100 nits. With no tone mapping, anything above 1.0 is cut off. ACES 2.0 SDR therefore puts 18% grey at 10 nits and scene 1.0 at about 46 nits, leaving room for highlights. Our sparks at about 62 are some 6 stops above diffuse white.
HDR displays have a much higher ceiling (the PQ signal goes up to 10,000 nits, while real displays vary by model). ACES 2.0 has a view for each peak: 500, 1000, 2000 and 4000 nits. Mid grey moves up only a little; most of the extra range goes to the highlights. Find the smallest view that keeps the sparks off the ceiling and gives diffuse white at least 100 nits.
peak above SDR (stops) = log2(peak nits / 100)2 Hands-on
Task: keep the sparks off the display ceiling (≤ 0.05% pinned) with diffuse white (1.0) at 100 nits or more
ACES 2.0 SDR saves most highlights, but check where diffuse white ends up, then step the HDR peak up from 500
3 Deep dive
From scene values to nits
Comp values are relative amounts of light (1.0 ≈ a normal diffuse white). Screens measure the light they emit in nits. Deciding how many nits 1.0 gets and where everything brighter goes is the output transform’s job, not the comp’s.
| ACES 2.0 view | 18% grey | Scene 1.0 | Sparks 62 |
|---|---|---|---|
| SDR 100 nits (Rec.709) | 10.0 nits | 45.8 nits | 99.1 nits |
| HDR 500 nits (P3 D65) | 13.2 nits | 89.1 nits | 475 nits |
| HDR 1000 nits (P3 D65) | 14.5 nits | 106.6 nits | 912 nits |
| HDR 2000 nits (P3 D65) | 15.8 nits | 121.7 nits | 1,707 nits |
| HDR 4000 nits (P3 D65) | 16.8 nits | 133.9 nits | 3,059 nits |
Sparks at 62 sit 6 stops above white; no SDR screen shows that straight
PQ and HLG
PQ (SMPTE ST 2084) is an absolute signal: a code value means a number of nits, up to 10,000. HLG is relative: each display scales it to its own peak. Both are part of ITU-R BT.2100.
Diffuse white in HDR
ACES 2.0 does not pin diffuse white to one number. Scene 1.0 lands at about 46 nits on SDR, 89 nits on HDR 500, 107 nits on HDR 1000 and 134 nits on HDR 4000: the higher the peak, the more the mid-tones rise, though only a little. ITU-R BT.2408 uses 203 nits as HDR reference white. The number you get depends on the delivery spec and the output transform.
What comp can do about it
What comp controls is the data: don’t clamp high values, don’t bake the view into files, and give highlights real detail (not flat white blobs), because an HDR screen will reveal what hides above 1. Check work through a view that rolls highlights off, or lower the Viewer exposure to look.
- 1Keep values above 1
- 2Check highlights at low exposure
- 3Deliver scene-linear
- 4Output transform per display
4 In real work
In Nuke
- Viewer: drop the exposure several stops to see whether highlights carry detail or are flat blobs; the image values don’t change.
- Viewer: pick an HDR display/view if your monitor, OS and config support it (check your version).
- Write (recap of CS4): hand-off EXRs stay scene-linear; viewable HDR files get the config’s HDR output transform.
- Clamp: only for building masks or mattes, never on the colour image you deliver.
Common mistakes → the fix
- ✗ Clamping at 1 before delivery “to be safe”✓ The HDR master loses its highlights. Deliver full values and let the output transform handle them.
- ✗ Highlights that are flat white blobs (for example a pre-clipped element)✓ They stand out on HDR screens. Add detail or use an unclipped element.
- ✗ Assuming every HDR display reaches 10,000 nits✓ 10,000 is the PQ signal’s ceiling; real displays are dimmer and vary by model.
Before you send the shot
- No Clamp on the colour image before the Write
- Highlights hold detail with the Viewer exposure lowered
- Hand-off EXRs are scene-linear (recap of CS4)
- You know the delivery spec: SDR, HDR or both
5 Check yourself
Why keep values above 1 in the EXR even for an SDR-only delivery?
🔒 Answer the question above and this card opens
1. A 1,000-nit HDR display: how many stops above the 100-nit SDR peak?
2. What is the main difference between PQ and HLG?
Where the numbers come from: Colour maths comes from src/figures/colormath.ts (sRGB IEC 61966-2-1, ARRI LogC3 EI 800, sRGB/P3/Rec.2020/ACES AP1 primaries and Bradford, checked against reference numbers). Real ACES 2.0 output via OCIO 2.6, studio config v4 (ACES studio config v4.0.0): SDR = sRGB - Display / ACES 2.0 - SDR 100 nits (Rec.709); HDR = Rec.2100-PQ - Display / ACES 2.0 - HDR 500/1000/2000/4000 nits (P3 D65). The neutral curves are exact OCIO samples every 1/16 stop; colour pixels go through a 33³ LUT baked from OCIO (within about 0.02 of exact). SDR clip is the Un-tone-mapped view cut off by the screen at 100 nits. Nits are decoded with PQ (SMPTE ST 2084, up to 10,000 nits). 18% grey = 10.0 nits (SDR), 13.2 / 14.5 / 15.8 / 16.8 nits (HDR 500/1000/2000/4000); scene 1.0 = 45.8 nits (SDR), 89.1 / 106.6 / 121.7 / 133.9 nits. ITU-R BT.2408 uses 203 nits as HDR reference white, but ACES 2.0 does not pin scene 1.0 to that number. “Pinned” = brightest channel at 98% of the peak or more. Measured: the goal passes with HDR 1000, 2000 and 4000.