1 The idea
A cube doesn’t say what its axes mean. With linear axes (amounts of light) our plate crowds near the black corner: half of its pixels are below about 0.14 on the grey axis.
Switch to sRGB or LogC3 coordinates and the same dots stretch upwards: the darks get more room. The cyan stop ladder shows how much room each encoding gives each stop.
2 Hands-on
Task: pick the coordinates that lift half of the plate’s pixels at least a third of the way up the grey axis
Try both sRGB and LogC3, then switch to Comp + sparks: which one keeps the sparks inside the cube?
3 Deep dive
The stop ladder in three coordinates
In linear each stop down gets half the room of the one above, so six stops under mid grey squeeze into the first 0.18 of the axis. sRGB stretches the darks, and LogC3 gives each stop about the same room (roughly 0.07 per stop around mid grey).
| Stops from mid grey | linear | sRGB | LogC3 |
|---|---|---|---|
| −4 | 0.011 | 0.108 | 0.153 |
| −2 | 0.045 | 0.235 | 0.257 |
| 0 | 0.180 | 0.461 | 0.391 |
| +1 | 0.360 | 0.634 | 0.463 |
| +2 | 0.720 | 0.865 | 0.536 |
Pick the coordinates for the job
Light maths (merge, blur, defocus, motion blur) belongs in linear, because it adds and averages amounts of light. Jobs where every stop should be treated alike, such as grading very bright fire or a blur that shouldn’t let hot pixels bloom, are often done in log and converted back.
- 1linear: light maths
- 2log: grade/filter HDR
- 3sRGB: only for viewing
The sparks show the ceiling
The hottest spark among the sampled dots is about 60 in red. In sRGB that is about 5.7, far outside the cube; in LogC3 it is about 1.01, almost exactly in the box. That is why log can hold such a wide range.
Same numbers, just a different ruler
4 In real work
In Nuke
- Colorspace: convert linear → sRGB or linear → AlexaV3LogC on a viewing branch or for a specific job.
- Log2Lin or OCIOColorSpace: always go into log and back out with a matching pair.
- PositionToPoints: feed it from different Colorspace settings to compare the clouds in the 3D Viewer.
- Viewer: sampled values are the graph’s values, whatever view the Viewer shows them through.
Common mistakes → the fix
- ✗ Blurring or defocusing an image converted to sRGB mid-graph✓ Do it in linear. If log is intended, convert in and out as a pair.
- ✗ Converting into log and forgetting to come back✓ Keep the in and out nodes together in one backdrop and check with a difference that the round trip returns the same values.
- ✗ Thinking a linear image that looks dark is wrong✓ Look through the right view; a cloud low in a linear cube is normal.
Before you send the shot
- Merges, blurs and defocus run in linear
- Every trip into log has a matching trip out
- No “for viewing” Colorspace left in the comp chain
5 Check yourself
In LogC3 coordinates, how is the stop ladder around mid grey spaced?
🔒 Answer the question above and this card opens
1. In linear, half the plate’s pixels sit below about 0.14. Is the plate too dark?
2. You want to soften a very hot explosion without the bright pixels blooming. How?
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). Height on the grey axis = the mean of the three encoded channels, measured on 25,680 plate pixels: median linear 0.140, sRGB 0.406, LogC3 0.363. The hottest spark value among those dots (~60 in red) encodes to about 1.01 in LogC3.