1 The idea
Diffusion works on the whole picture: blur it and mix it in low, so light only moves around and the frame average barely changes. Glow works on bright lights only: key the highlights, blur them at several sizes, add. Veiling glare is light scattered all over the lens: a thin haze that lifts every black. Watch the graph and the “darkest 1%” number to see where each one acts.
A glow ladder (blur in several steps, add them all) is often not normalised: if the steps’ weights add up past 1, the energy shoots up. The check: feed it a flat colour (Constant 0.5) and measure what the glow sends out. This lab starts at ×5.9, the same as one glow tool measured in Nuke 17.
glow = brightness · Σ spreadᵏ · blurₖ(highlights) · flat field: out ÷ in = brightness · Σ spreadᵏ2 Hands-on
Task: set the brightness so the flat field comes out ×1.00, then set the mix to add 5–30%
The normalising brightness = 1 ÷ Σ spreadᵏ (the readout shows it). The Flat-field test view helps.
3 Deep dive
Three things often all called “glow”
All three make the picture softer and more luminous, but they act on different parts of the picture and do different things to the light’s energy. Read which one the plate has, and add that one.
| Acts on | Energy | You see | |
|---|---|---|---|
| Diffusion | The whole image | Moved; the average barely changes | Soft overall, blacks blend |
| Glow / bloom | Highlights only | Added | Bloom around lights |
| Veiling glare | Light scattered across the lens | Added, spread over the frame | Lifted blacks, lower contrast, less saturation |
The glow ladder, and why it blows up
A glow that looks like a real lens isn’t one blur size but many stacked, because real lens scatter has a long tail. So we build a ladder: blurs that double in size, each multiplied by a weight, all added. With weights of spreadᵏ and a spread above 1, the widest step weighs the most, giving a broad, foggy glow.
The sum of the weights is the glow’s energy multiplier. Ten steps at spread 1.12 add up to 19.65; times a brightness of 0.3 that’s 5.9×. On a flat image the glow sends out almost six times the light it was given.
The flat-field test
A normalised blur must return the same value on a flat image, whatever its size. That makes a flat image the measuring tool: feed a Constant 0.5 with the threshold off and sample only what the glow itself makes (turn on effect only, or subtract the input). 0.5 means normalised; more means the glow quietly adds light of its own. Without effect only, Nuke’s Glow adds its result onto the input, so a normalised one reads 1.0, not 0.5.
Once it’s normalised, the mix or intensity means exactly what it says: a mix of 0.2 adds 20% of the keyed highlights. Dial it to what looks right, then try half; it’s often still enough.
- 1Constant 0.5
- 2Glow (threshold off, effect only)
- 3Sample the value
- 4Adjust until 0.5
- 5Control with mix
Measure every glow gizmo on a Constant before trusting its default
Why one main glow, after the merge
A real lens scatters the whole picture at once, with one character. Glow each element separately and each gets its own size and energy, stacking wherever elements overlap. One normalised glow after the CG and plate are merged, at the head of the camera stack (with diffusion, before the flares and the shake; see LA7), makes every highlight in the frame bloom the same way. An extreme light, such as an explosion flash, can get its own controlled glow if needed.
4 In real work
In Nuke
- Test: Constant 0.5 → the glow you plan to use (threshold off, or below 0.5, and only the glow’s own part: effect only, or the input subtracted) → sample in the Viewer or with CurveTool. Expect 0.5.
- A hand-built ladder: several Blurs doubling in size → Multiply each by its weight → Merge plus them together → multiply by 1 ÷ the sum of the weights.
- Nuke’s Glow node: use effect only to get just the added part, then Merge plus it yourself with a mix.
- Diffusion: a very large Blur → Merge (over, or a mix) at a very low mix; check with viewer gain.
Common mistakes → the fix
- ✗ Trusting a glow gizmo’s defaults✓ Measure it on a Constant first; if it’s above ×1, normalise or grade it down.
- ✗ A glow per element, all stacking✓ One main glow after the merge; split off only the extreme lights.
- ✗ Using diffusion where glow belongs (or the other way round)✓ Read the plate: soft all over = diffusion, bloom only around lights = glow, lifted blacks everywhere = veiling glare.
- ✗ A glow so strong the image washes out✓ Normalise, control the mix, then try half of what looked right.
Before you send the shot
- The glow passed the flat-field test (×1)
- One main glow after the merge
- Blacks around lights lift only slightly; check with viewer gain
- The kind of softness (diffusion / glow / veiling) matches the plate
5 Check yourself
A glow fed a Constant 0.5 outputs 2.95 in its own part (effect only). What does that mean?
🔒 Answer the question above and this card opens
1. A 4-step glow ladder has weights 1, 1, 1, 1. What brightness normalises it?
2. The plate has hazy, lifted blacks everywhere and low contrast, but no clear bloom around the lights. What should the CG get?
Where the numbers come from: The ×5.9 (0.3 × Σ 1.12ᵏ, k = 1…10 = 0.3 × 19.65) matches what KB 159 measured in Nuke 17 on one glow gizmo. The flat-field test here really runs the same blurs on a flat image (no key). Blurs are three box passes at reduced resolution. The diffusion mix and veiling-glare amounts are teaching values; real shots use far less and are checked with viewer gain. Veiling glare is approximated here by a very large blur.