1 The idea
A shot that looks fine at normal settings can hide problems that only show with the viewer pushed: mismatched blacks, paint patches, halos. That’s why compositors QC with gain and gamma before every delivery.
This image hides two faults. Push gain and gamma, then click on them.
2 Hands-on
3 Deep dive
Push the viewer, not the comp
At normal exposure, problems in the darks and highlights stay hidden, then show up the moment the shot reaches the colour grade. Viewer gain, gamma and mirror never reach the render, so you can push them hard without risk.
Run the same pass every time, one step after another, so nothing gets skipped.
- 1Gain up
- 2Gamma up
- 3Gain down
- 4Saturation ×9
- 5Mirror / wipe
- 6Cycle backgrounds
What each test finds
No single test finds everything. Some faults show only with gain up and others only with gain down, so do both.
| Test | What it finds |
|---|---|
| Gain / gamma up | Blacks off per zone, dark or bright rims, leaking roto |
| Gain down | Clipped values above 1, halos in bright areas |
| Saturation ×9 | Spill, colour fringes, CG colours too strong for the scene |
| Mirror | Fresh eyes on things you’ve stopped seeing |
| Difference vs plate | Places you changed the plate by accident |
Wipe and difference: 0 outside the work area
Wiping the comp against the original plate is the most honest check. Then Merge the comp and plate with difference and push the viewer gain way up: everything outside the area you meant to change must be pure black. If you comped on a denoised plate, diff after the step that puts the original plate back outside the elements (I4); before it, the whole frame differs by the grain.
Grey specks in the sky far from the robot mean something touched the plate there, such as a grade with no mask, or an element with rgb values where its alpha is 0.
Cycle the background under the element
Put the element over a white Constant, over a 3.0 Constant, and over a checker. Dark fringes jump out on white or bright backgrounds but can hide on a checker; holes in the alpha show on the checker.
Finish with the technical checks: NaN or inf pixels and negatives that shouldn’t be there. A big Blur helps, because it spreads NaNs until they’re easy to see.
The hidden fault is usually in the test you were about to skip.
4 In real work
In Nuke
- Viewer: slide gain up and down, and gamma up.
- Input Process: make a ColorCorrect (saturation around 9) or a Mirror, name it VIEWER_INPUT, and toggle it with the viewer’s IP button.
- Wipe: plug comp and plate into two viewer inputs and turn on wipe.
- Merge (difference) the comp against the plate and push the viewer gain.
- Constant (white, 3.0) and CheckerBoard → Merge (over) under the element, one at a time.
Common mistakes → the fix
- ✗ Checking only at normal exposure✓ Always do gain up, gamma up and gain down.
- ✗ Adding a Grade in the tree to inspect, then forgetting it✓ Use the viewer or Input Process; they never reach the render.
- ✗ Letting grey in the difference outside the work area slide✓ Track down the node touching the plate until outside the work area is 0.
- ✗ Checking the element only over a checker✓ Also cycle to white and bright backgrounds, where dark fringes show.
Before you send the shot
- Gain up, gamma up, gain down: no wrong edges or blacks.
- Saturation ×9: no spill or colour fringes.
- Difference vs plate is 0 outside the work area.
- Over white, 3.0 and checker: no fringes or holes.
- No NaN, inf or stray negatives.
- It still holds up mirrored.
5 Check yourself
To find mismatched blacks, how do you push the viewer?
🔒 Answer the question above and this card opens
1. The comp-vs-plate difference shows grey specks in the sky, far from the robot. What does it mean?
2. You want to find a dark rim around the robot. Which test shows it best?
Where the numbers come from: Computed in your browser: viewer gain and gamma applied to a Blender render of our desert scene with two planted faults. A teaching viewer, not Nuke’s.