The homepage tips carry the short version — middle grey at 41% IRE, Zebra 1 at 94%, skin in the 55–62% band. This guide is where those numbers come from, and what the curve does either side of them. Those three targets are the site's; every other figure below is computed from the transform library this site ships, so the arithmetic is yours to check rather than take on trust.
What S-Log3 encodes
S-Log3 is a recording format, not a viewing one: a rule saying which scene luminance gets which number in the file. This site ships that rule as executable code, so the anchor points below are read out of the curve rather than remembered from a forum post.
| Scene value | S-Log3 code value |
|---|---|
| Zero reflectance (absolute black) | 9.3% |
| 4 stops under middle grey | 16.7% |
| Middle grey, 18% reflectance | 41.1% |
| 90% white card | 58.4% |
| Ceiling — code value 100% | linear 38.4, i.e. 213× middle grey, 7.74 stops over |
The curve is a fixed function of scene reflectance, with no exposure or gain term anywhere in it: 18% reflectance lands on 41.1% by definition of the format. That is why 41% is published as a flat target rather than something to recompute per shot.
Notice how the container is divided. Above 41.1% there are 58.9 points of code value, covering 7.74 stops of highlight. Below it there are 24.3 points down to the toe — four stops — and 7.5 more points of compressed floor beneath that. Of the range the format actually uses, 9.3% up to 100%, nearly two thirds sits above middle grey.
A waveform reading is a stop count in disguise
The S-Log3 log branch carries a small linear offset inside the logarithm, so a stop is worth fewer code-value points near the bottom of the curve than near the top.
| Stops from middle grey | Code value | Points to the next stop |
|---|---|---|
| −4 (toe) | 16.7% | 4.7 |
| −2 | 27.3% | 6.6 |
| −1 | 33.9% | 7.1 |
| 0 | 41.1% | 7.4 |
| +1 | 48.5% | 7.5 |
| +2 | 56.0% | 7.6 |
| +5 | 79.0% | 7.7 |
| +7 | 94.3% | — |
Above middle grey the spacing is near enough constant to use in your head: subtract 41, divide by 7.7, and you have your stops over. From middle grey up to clipping, that shortcut lands within a tenth of a stop of the real figure. Below middle grey it drifts — at the −2 row above, 27.3%, it reports 1.8 stops under when the true answer is 2 — so it is a highlight tool only.
Below four stops under grey, the curve stops being logarithmic
At exactly 0.18 ÷ 16 the curve hands over to a straight line running down to the 9.3% floor. Everything darker than four stops under middle grey — not four stops' worth, but everything — is squeezed into the band between 9.3% and 16.7%: 76 code values out of 1,023. For comparison, the single stop immediately above middle grey gets 76 code values to itself, and the widest stop in the highlights gets just under 79.
Because that bottom segment is straight rather than logarithmic, each stop down halves what is left of it. From four to five stops under grey you get 38 code values; the next stop gets 19; the one after that, 10. That is the real cost of under-exposing, and it is not "a bit of shadow detail" — it is moving your subject out of a region where one stop has 76 values of its own into one where a stop is worth 38, then 19, then 10.
How much room sits above the zebra
The arithmetic is exact. Code value 100% is 7.74 stops over middle grey; the 94% zebra threshold sits at 6.96 stops over. So the zebra fires roughly 0.78 of a stop before the file runs out. It is an alarm with a little under a stop of warning in it, not a clipping indicator.
Expose skin into the 55–62% band and you are 1.9–2.8 stops over middle grey, leaving 5.0 to 5.9 stops of room above the face. That band already assumes about a stop of exposing to the right, which is what keeps noise out of the shadows; at a strictly nominal exposure skin sits lower.
Not all of that room survives conversion. This site's transform chain white-points at linear 8.0 — 5.47 stops over middle grey — and everything above that lands on white. Through this chain, the top ~2.3 stops S-Log3 records are grading room, not picture.
Range clipping and gamut clipping are different failures
Two things both get called clipping, and only one of them is an exposure problem.
Range clipping is a channel reaching code value 100%. Exposure fixes it. Gamut clipping is a colour that has no Rec.709 coordinate at all. Pulling exposure down does not touch it.
The second is visible in this site's transform library. The S-Gamut3.Cine → Rec.709 matrix, applied to a pure blue at the camera's primary, gives (−0.087, −0.228, 1.399): two negative channels. Pure green gives (−0.540, 1.341, −0.472). Negative means outside the container, and the shipped chain clamps those channels to zero because there is nothing else it can do with them.
So on a saturated blue LED wall, the two clamped channels go flat across the whole wall, and every difference the wall was carrying in them — the gradient, the fall-off at the edge of each fixture, the spill on a shoulder — goes with it. In the grade the instinct is to pull the wall down. It does not help; it was never a level problem.
What to do instead:
- Handle it at the Color Space Transform. Set Gamut Mapping Method to Saturation Compression and use Saturation Knee to place where the roll-off begins. That compresses the extreme LED and laser saturations back inside the container smoothly instead of letting them clamp and posterise.
- Do not reach for Highlight Recovery. It is a Camera RAW control, it applies to RAW formats such as X-OCN and BRAW, and it does nothing for XAVC S-Log3.
Where this exposure lands in the grade
Correct your exposure on the native log image, convert, then apply the look. Offset only behaves like a real exposure adjustment on log data, and once the transform has compressed your highlights into Rec.709's smaller range, pulling exposure down drags that block as a whole rather than recovering detail.
For the conversion itself, use a Color Space Transform or one of the log to Rec.709 conversions — the Sony S-Log3 conversion here is one of six baked at 33×33×33 from the same transform library the numbers above come from. The DaVinci Resolve page has the node order in full, and concert LUTs covers the gamut side of stage lighting from the grading end.
Shooting X-OCN RAW instead? Node 1 becomes the Camera RAW tab, and the node tree is in grading FX5 X-OCN RAW in Resolve.
Sources
- The transform library this site ships — the S-Log3 decode and encode curve, the S-Gamut3.Cine → Rec.709 matrix, the shared shoulder that white-points at linear 8.0, and the clamp on negative channels. Every anchor value, stop-spacing figure and matrix result above is computed from this file.
- The site's colour-pipeline notes — the six 33×33×33 conversion LUTs, and 18% grey at 0.4106 for S-Log3.
- The site's grading notes on traps — correct before you convert; Offset behaving like exposure on log data only; Highlight Recovery being a Camera RAW control; Gamut Mapping → Saturation Compression as the control for blown stage colour.
- The homepage tips and FAQ — the 41% / 94% / 55–62% figures, the note that the skin band assumes roughly a stop of exposing to the right, and S-Log3 as a recording rather than a viewing format.
- The concert and DaVinci Resolve LUT pages — out-of-gamut stage LED and why pulling exposure down does not fix it, and the three-node order this guide links to.