All posts
By Pravit Gandhi··9 min read

S-Log3 to Rec.709: LUT vs color space transform

Should you convert S-Log3 to Rec.709 with a LUT or a color space transform? Here is what each one actually does, what it breaks, and the order to work in.

A LUT and a color space transform are two ways to do the same technical job: take S-Log3 encoded footage and put it into Rec.709 so it can be viewed and delivered. A LUT is a fixed lattice of pre-computed values that the software samples and interpolates between. A color space transform computes the conversion on the fly, inverting the log curve, converting the primaries and applying the output curve. Both are normalization. Neither is a grade, and the most expensive mistake in log workflows is treating them as one.

What a technical conversion actually is

Converting S-Log3 to Rec.709 means two independent operations, and confusing them is where most bad conversions start.

The first is the transfer function. S-Log3 is a log encoding curve that Sony built on the Cineon digital negative revised in 2007, with no shoulder and a reduced toe. Sony's technical summary places 18% gray at 41% (10-bit code value 420) and 90% white at 61% (code value 598). Rec.709 puts those in very different places. Undoing one curve and applying the other is the tonal half of the job.

The second is the gamut. S-Gamut3.Cine, the container most people shoot alongside S-Log3, is described by Sony as similar to a negative film scan and designed slightly wider than DCI-P3. S-Gamut3 is wider still and close to camera native, which is why Sony recommends it for ACES and BT.2020 work and for archiving. Rec.709 is much smaller than either. Squeezing the larger gamut into the smaller one is the color half of the job, and it is the half a badly built LUT gets wrong.

A technical conversion is the transform that makes log footage viewable and nothing else. Every decision about contrast, warmth, skin and mood happens after it.

The official LUTs, and what is actually inside them

Sony supplies conversion LUTs. Its technical summary confirms the company provides .cube Look Profiles to convert S-Gamut3.Cine/S-Log3 into 709 color space, and answers the direct question about 3D LUTs for 709 with a straightforward yes.

What is worth understanding is that Sony's own 709 options are not neutral, and Sony says so. Its log shooting guide describes s709 as producing "film-like color tones and color reproduction, with low contrast," and 709(800%) as producing "color reproduction and high contrast suitable for TV broadcasting." Those are two different looks with a technical conversion inside each. Picking one over the other is already a creative decision that you have delegated.

The same thing is happening inside your editor. Apple's Final Cut Pro applies its Sony S-Log3/S-Gamut3 and S-Log3/S-Gamut3.Cine camera LUTs as "the appropriate linearization and 709(800%) MLUT tone mapping," which is Apple naming the exact look it has chosen on your behalf. That is a reasonable default and it is not a neutral one.

Opinion: use the manufacturer or NLE transform as your starting point and know which one you picked, rather than downloading a conversion LUT from a preset pack. Free packs bundle a technical conversion with someone's creative bias, and once it is baked into a .cube you cannot separate them.

Where a LUT and a CST genuinely differ

A LUT is a lookup table. It stores output values for a grid of sampled input points and interpolates everything in between. Feed it an input outside the range it was built for and behaviour becomes implementation dependent: clamped, extrapolated, or simply wrong.

A color space transform is arithmetic. It inverts the log encoding to linear, applies a matrix to move between the source and destination primaries, then applies the destination transfer function. There is no sampled grid, so there is no interpolation error and no edge of the table to fall off.

The consequences show up in specific places. LUT interpolation error concentrates where the sampled grid is coarse relative to how fast the values are changing, which in practice means saturated highlights: a sunset that skews magenta, an LED practical that shifts hue as it brightens, a neon sign that goes a color it never was. A CST does not produce that class of error, but it will happily clip anything outside the destination range unless you turn tone mapping on, which is the mistake people make in the opposite direction.

Neither is universally correct. Use a LUT when you need portability across applications and machines, because a .cube file loads anywhere. Use a CST when you are mixing camera formats, when exposure varies across the timeline, or when you need to know precisely what the pipeline is doing.

The range trap that eats your blacks

This one is worth slowing down for, because it produces symptoms that look like grading problems and are not.

Sony records S-Log3 as full range. Its technical summary is explicit: data runs from 0 to 1023 in 10-bit, which corresponds to minus 7 IRE through 109 IRE, and camera SDI output is full range every time with no option to select legal range. Legal range by contrast is scaled and limited to code values 64 through 940, or 0 to 100 IRE.

So if any step in your chain assumes legal range data, everything outside 64 to 940 is discarded and the levels shift underneath you. The symptom is crushed blacks that no lift recovers and highlights that clip earlier than they should, and it survives every creative adjustment you make afterwards because the damage happened before your first node.

The check takes ten seconds. Put a waveform on the untouched clip and confirm that an 18% gray card reads 41, or code value 420 if your scope is showing raw 10-bit. If it does not, fix the interpretation before you fix anything else. This is also why the same file can look correct in one application and wrong in another with identical settings.

The order of operations that keeps you out of trouble

Normalize, balance, then look. In that order, always.

Normalize means the technical conversion, whether that is a CST node, an NLE camera LUT, or a manufacturer .cube applied as an input transform. Balance means fixing exposure and white balance so the shot sits where it should before any stylistic decision. Look means the creative grade, and delivery transform last if your pipeline uses one.

The reason the middle step comes before the look and not after is specific. A conversion LUT is built assuming correct exposure. Feed it a clip that is two stops hot and the bright values map through the top of its curve, where the tone response steepens and hue shifts accelerate. Skin is what shows it: faces go orange as they brighten and the correction you apply afterwards cannot unpick it, because the skew happened inside the lookup table. Where you put that exposure correction depends on which route you took. With a conversion LUT, the offset has to sit upstream of it, while the clip is still in log, because once values have been mapped through the table the skew is baked in. With a color space transform you have the easier option: put the input transform first, then correct exposure in the working space that follows it. Either way the correction happens before the creative look, never after. Getting exposure right in camera avoids it entirely, which is covered in how to expose S-Log3 correctly.

In a node-based pipeline that means an input transform first, your grading nodes in the working space, then the output transform. In Premiere it means letting color management handle the conversion and grading in Lumetri on top, without a conversion LUT in the Creative tab. In Final Cut it means setting the Camera LUT in the Info inspector rather than dropping a Custom LUT effect on the clip, since the camera LUT is applied before your adjustments and the effect after. None of this requires a node graph, which is the point of grading log footage without DaVinci Resolve and of the wider set of Resolve alternatives that now handle log natively.

Leumos AI, our browser-based grading studio, treats this as the fixed part of the pipeline: S-Log3 uploads get an input transform applied before anything else, so every shot on the timeline is already normalized when the look goes on. It is in closed beta with a waitlist. The principle is tool independent and worth stealing regardless of what you grade in, and it is the same reason automated tools work at all, which we unpacked in what AI color grading actually means.

Artifacts, and how to tell them apart

Four failure modes account for most bad S-Log3 conversions, and each one has a signature.

Hue skew in saturated highlights means LUT interpolation. Look at the brightest colored objects in frame, not the skin. If a sunset or a practical is drifting toward a hue that was not in the scene, swap the LUT for a computed transform and see if it resolves.

Banding in smooth gradients means you stretched too few code values. S-Log3 puts 325 of its 1023 levels below middle gray, so a heavy contrast curve applied instead of a proper transform produces steps in skies and walls. In 8-bit source material the same span is about 81 levels and the banding is unavoidable rather than merely likely.

Crushed blacks and early clipping usually mean a range mismatch or a CST with tone mapping disabled. Check the waveform on the raw clip first, then check whether your transform has a tone mapping option you left off.

Plastic highlights and skin that goes orange as it brightens mean two transforms are stacked. Something is converting your log twice: automatic color management plus a manual LUT, or a camera LUT plus a LUT effect. Find the second one and remove it. This is the most common single fault in Premiere projects and it comes from a well-intentioned feature doing its job while you also do it, which is one of the symptoms in why your S-Log3 footage looks washed out.

Once the conversion is clean, matching shots to each other is a separate exercise, and a much easier one, since every clip now starts from the same defined space. How much easier has been measured: across 21 format pairs, correctly normalized log encodings disagree by nothing at all. That is the premise behind automated shot matching and it falls apart if the normalization was inconsistent to begin with.

Frequently asked questions

Is a LUT or a color space transform better for S-Log3 to Rec.709?

A CST is more accurate and a LUT is more portable. The transform computes the conversion and has no interpolation error or table edge, while a .cube file loads in any application on any machine. For a single camera project delivered from one editor either is fine. For mixed formats or variable exposure, use the transform.

Why does my skin go orange after applying an S-Log3 to Rec.709 LUT?

Usually because the clip is overexposed relative to what the LUT expects, so the bright values map through the steepest part of its curve. Correct exposure before the conversion rather than after. If exposure is nominal and it still happens, you have two transforms stacked, or a LUT with a creative look baked in.

Do I need to use S-Gamut3.Cine or S-Gamut3 with S-Log3?

Sony describes S-Gamut3.Cine as slightly wider than DCI-P3 and easier to grade, and S-Gamut3 as close to camera native and better suited to ACES, BT.2020 and archiving. For Rec.709 delivery, S-Gamut3.Cine is the more convenient container. What matters most is that you declare the right one in your transform, since the two are not interchangeable.

What order should I apply the conversion and the grade?

Conversion first, exposure and white balance second, creative look third. Applying the conversion last means every adjustment you made was made on log values and will not behave as you expect once the transform runs. Applying the look before the conversion means the LUT reinterprets your creative decisions.

Sources