False Color Before You Touch the Grade
A 30-second false-color pass that flags clipped highlights and muddy faces before any LUT, with the IRE bands to watch on generated skin and chrome.
Operators skip exposure because the thumbnail looks fine. Generated frames are lit to please a still: faces are smooth, products are bright, skies are even. False color is a 30-second overlay that ignores all of that and maps luma to a handful of bands. Clipped chrome goes to the clip colour. A muddy face sits in the near-black band. You see it before a LUT stretches either mistake into something you cannot undo.
Do this pass before any look, before any show LUT, before any "cinematic" cube. A creative LUT will not fix mismatched shots, and it will not restore a highlight that is already a rail.
The thumbnail lies about exposure
A generated talking head can look "well exposed" in a 320-pixel poster and still be a clip on the waveform. The sampler likes bright, even keys. Product chrome picks up speculars that have no texture, only a white slab. Darker skin often lands too low because the model treated "cinematic" as "pull the mids down." None of that is obvious in a recap grid.
False color is not a standard with one set of colours. ARRI, RED, Sony, and Resolve each map IRE (or code value) to different overlays. Do not memorise someone else's pink as gospel. Memorise the bands, then learn which overlay colour your tool uses for each band. On Rec.709 legal range, 0 IRE is reference black (code 16 in 8-bit) and 100 IRE is reference white (code 235). Middle grey, 18% reflectance through the BT.709 OETF, sits a little over 40 IRE. That mapping is the scope. False color is a paint job on top of it.
You need a tool that will apply the overlay. The video editor is an EDL-based timeline for assembly; it is not a false-color scope. Work the file in the grading app that has the overlay. Do not QC exposure on the editor's preview: true pass. That pass is a free 480p render with a short per-user cooldown, billed as described in editor previews and the final export. At 480p, clip is harder to see and so is mud.
If the file is full-range RGB tagged as Rec.709, false color set to legal will lie: blacks look super-black and whites look super-white because the codes really are 0 and 255. Probe range first, or you will regenerate plates that only needed a convert.
A 30-second false-color pass
Time it. The point is that it is cheaper than a grade, and cheaper than discovering clip after a LUT.
0-5 seconds: overlay on, look off. Disable every LUT, every automatic match, every "reframe and enhance." You want the file.
5-15 seconds: highlights. Find the brightest thing that is supposed to have texture: chrome, a window, a practical, a white pack. If that region is a solid clip colour (whatever your tool uses for 100 IRE and above) with no holes of the band just below, it is a rail. A real specular is a small peak: clip colour in the core, then a falloff through the high bands. Generated chrome often has no falloff. That plate will not take a look LUT. It will take a regenerate with a dimmer key and a prompt that names the light instead of asking for "bright studio product lighting."
15-25 seconds: faces. A face should not sit in the near-black band and should not sit in the near-clip bands. Generated lighter skin often sits too high, in the band you wanted for a white shirt. Generated darker skin often sits too low, in a band that will turn to mud as soon as you add contrast. You are not matching a magazine IRE for "correct" skin. You are asking: is there room to grade, or is the face already against a rail?
25-30 seconds: the product and the backdrop. A labelled pack should not be clip-colour on the white of the card. A sky should not be a stack of two false-color bands with a hard edge between them; that is banding you will meet later. If the product is the reason the shot exists, the product's texture wins over a pretty face.
Pass/fail in one line:
| Region | Pass | Fail, regenerate |
|---|---|---|
| Specular (chrome, glass, window) | Small core at high IRE, visible falloff | Flat clip-colour slab, no falloff |
| Face | Occupies a mid band with room above and below | Against the floor or against the ceiling |
| Product card / white pack | High, not a rail; type still separate | White of the pack is the same band as clip |
| Backdrop / sky | A ramp across bands, not two flats | Two bands with a hard join (already posterized) |
If two takes both pass, pick on content, not on which thumbnail was brighter. If none pass, change the lighting sentence and generate again from text to video. Do not "fix" a fail with offset in the grade. What a grade can and cannot rescue includes this case: a rail has no values to redistribute.
IRE bands on generated skin and chrome
Hold a consistent band across shots of the same face. The absolute number matters less than the agreement, because generated skin is not a measured reflectance.
- Below ~10 IRE. No recoverable shadow. If a face is in this band, the key was too weak or "moody" in the prompt ate the mids.
- Around 40 IRE. Middle grey neighbourhood on Rec.709. A grey card or wall, not a skin target.
- Faces. Lighter generated skin often lands 55-70 IRE when the sampler is trying to be pretty; darker generated skin often lands 35-50 IRE when it is trying to be dramatic. Neither is a law. The fail is a face that shares a band with clip, or with crushed black. Pick a band for this character and hold it on the reverse.
- 90-100 IRE. Speculars and white props. A peak that kisses 100 IRE is fine. A region that lives there is clip waiting for a LUT to advertise it.
- Above 100 IRE on a legal-scale overlay means super-white or full-range codes. Either you have headroom to legalize later, or you have a range tag problem. Do not grade it as a look.
Chrome is the tell on product work. Real chrome has a small, hard specular and a falloff that carries the shape of the source. Generated chrome often has a white patch the size of the whole highlight: one false-color band, edge to edge. You cannot recover a shape that was never in the codes. Reroll with a larger, dimmer source ("soft window light, no hard sun on the metal").
Skin is the tell on talking heads. If false color shows the forehead in a near-clip band and the eye sockets in a near-black band, the sampler invented a key that does not exist in the room. A grade that "evens it out" flattens the face into plastic. Regenerating with a named key and fill, the same sentence on every take, is the lighting continuity fix. False color is how you notice you need it in thirty seconds instead of after a LUT.
Write the fail into the shot log: 03-b clip chrome, 05-a face in floor band. A defect taxonomy for generated video would file a clipped highlight as RENDER if it is visible in one frame, which it is. Do not file it as a grade note. It will sit in a colourist's queue for an hour and come back as the same rail in a different LUT.
What to do when the overlay is screaming
Regenerate first when the fail is a rail. Change one thing in the prompt: source size, source brightness, or the adjective that was actually a grade ("cinematic," "dramatic," "high contrast"). Keep the framing. A new angle on a still-clipped product is a new clip of the same defect.
If the overlay is noisy (bands flickering on a face that is otherwise mid), you may be looking at compression, not exposure. Pause on a still. If the still is clean and the overlay crawls in motion, that is a bit-budget problem on a gradient, not an IRE problem on a face. Hand that clip to the banding checklist, not to an exposure offset.
If only one shot in a matched pair fails, regenerate that shot on the same model as the pass. Mixing a dim take from family A with a bright take from family B to "average" them is how you end up with two CDLs and no show.
Once every hero take passes, turn the overlay off and grade: exposure (you should need very little), balance, sat, look.
FAQ
Which false-color LUT should I use?
The one in your grading tool that is documented, with a legend you can see. Colours are not portable across vendors. IRE bands are. If the tool has a Rec.709 legal false color and a "cinema camera" false color, use Rec.709 legal for generated SDR. You are not on a LogC set.
Can I leave false color on while I grade?
No. It is a QC overlay. Grade on the picture and on waveform, parade, and vectorscope. False color will fight you: you will start grading the overlay. Thirty seconds on, then off.
Does this replace a waveform?
No. False color is a fast spatial view of the same luma the waveform plots. Use false color to find the region, waveform to measure it. A face that is half clip-colour is obvious in false color and easy to miss on a waveform that also contains a window.
What if the thumbnail is dark on purpose?
Then the face should still occupy a band with room to move, and the blacks should not be a solid floor-colour with no holes. Intentional underexposure is a small key and held shadow detail, not a crushed code-16 slab. If the overlay is a single floor colour across the frame, there is no look in there to recover.