The compositor’s relationship with color space is one of those areas where the textbook answer and the practical answer diverge in ways that trip up artists who’ve only learned one or the other. The textbook says: “Composite in linear, because that’s how light actually behaves. Display in whatever the deliverable requires.” Which is correct. The practical reality says: “Sometimes you’re handed log footage, sometimes you’re handed Rec.709, sometimes you’re handed ACES, and the right working color space depends on the shot, the pipeline, and what’s downstream.”
Both are right. The textbook describes the principle; the practice describes the application. A compositor who only knows the textbook fights the pipeline; a compositor who only knows the practice produces shots that don’t survive review. The craft is knowing both.
This post is the working compositor’s guide to color space decisions. What linear, log, and ACES actually do; when each is the right working space; and how to decide on a per-shot basis instead of dogmatically.
Why Linear Is the Correct Color Space for Compositing
Linear color values represent light intensity directly. A pixel value of 1.0 corresponds to twice as much light as a pixel value of 0.5. This linearity is what makes math behave correctly: addition combines light sources properly, multiplication models attenuation properly, and the operators that compositors use most often (Over, In, Out, Atop, the Vlahos color difference method, premultiplied alpha math) all produce correct results.
Non-linear color spaces — log, gamma-corrected sRGB, Rec.709 — do not behave this way. A Rec.709 pixel at 0.5 doesn’t represent half the light of one at 1.0; it represents some perceptually meaningful midpoint that depends on the gamma curve. Math performed in non-linear space produces results that look subtly wrong: highlights that don’t blow out the way they should, soft edges that read as gray instead of bright, color combinations that drift toward muddy.
The textbook reason to composite in linear is to get correct math. The practical effect is fewer “fix it in the next pass” problems caused by edge cases that linear handles correctly and non-linear doesn’t. Compositors who do their work in linear find that aggressive operations — strong color grading, high-contrast composites, blowout effects, exposure compensation — produce more predictable results.
What Log Is For
Log color space exists because real cameras have to capture far more dynamic range than a Rec.709 display can show. A modern cinema camera captures 14+ stops of dynamic range. Rec.709 displays cover roughly 6–7 stops. Recording the camera’s full range as a Rec.709 image throws away most of what the camera saw — highlights clip, shadows crush, the recorded image is a small slice of what the sensor actually captured.
Log encodings preserve the full range by mapping it logarithmically. The brightest pixels and the darkest pixels both fit in a single 10-bit or 16-bit file, with detail preserved in both directions. Visually, log footage looks flat and grayish because it’s compressing a huge range into a viewable file — but the range is all there, ready to be graded out in post.
What this means for compositing: log footage has more information than Rec.709, and that information matters when the shot has to handle exposure changes, highlight extensions, shadow detail, or aggressive color treatment. Compositing in log lets you preserve that information through the work.
But — and this is where the textbook answer gets nuanced — compositing math doesn’t behave correctly in log. The log curve is non-linear, so the operators that work correctly in linear produce subtle errors in log. The standard practice is to convert log to linear at the start of the comp, work in linear, and convert back to log for delivery (or to whatever the deliverable color space is).
ACES: The Pipeline That Bridges Both
ACES — the Academy Color Encoding System — is the modern industry-standard color pipeline that handles the linear-vs-log problem cleanly. The structure is:
- Input device transform (IDT): takes camera-native log footage and converts it to ACES working space, which is a wide-gamut linear space.
- Reference rendering transform (RRT) and output device transform (ODT): takes ACES working space and produces the deliverable in whatever color space is needed (Rec.709 for HD broadcast, P3 for theatrical, sRGB for web, HDR for streaming).
The compositor working in an ACES pipeline composites in linear (the working space), with the IDT handling the conversion from camera-native log on the way in and the ODT handling the conversion to deliverable on the way out. The math is correct, the highlights and shadows are preserved, and the deliverable is graded properly for its target medium.
The benefit of ACES isn’t just technical correctness. It’s pipeline consistency across vendors and across projects. A shot worked in ACES at one studio plugs into a shot worked in ACES at another studio without color management surprises. The IDT and ODT for each camera and each deliverable are standardized; the working space is identical regardless of what was shot or what’s being delivered.
For productions running ACES, the compositor doesn’t make color space decisions on a per-shot basis. The pipeline makes the decisions; the compositor works in linear ACES and trusts the IDT/ODT structure to handle the conversions.
When You’re Not in an ACES Pipeline
Plenty of productions don’t run ACES. Smaller commercial work, regional markets, productions where the DI house uses a different pipeline — all of these end up with non-ACES color management. In those cases, the compositor has to make color space decisions per shot.
The basic decision tree:
- What color space was the plate captured in? Camera log usually (S-Log, Log C, V-Log, RedLogFilm, etc.). Sometimes Rec.709 if the production was budget-constrained or used consumer-grade cameras.
- What’s the working space the comp will be done in? Linear, almost always — unless there’s a specific reason not to.
- What’s the deliverable color space? Rec.709 for HD broadcast, P3 for theatrical, sRGB for web, often log for archive, sometimes ACES for next-stage handoff.
- What conversions are needed at each stage? Plate → working space → deliverable. Each conversion is a place where information can be lost; the goal is to minimize the number of conversions and make sure each one is correct.
In practice: convert log to linear at the start, composite in linear, convert linear to deliverable color space at the end. The two conversions are the standard pipeline. Trying to skip one (compositing directly in log, for example) usually produces visible errors in the final shot.
The Premultiplied Alpha Question
A related issue that compositors run into around color space is premultiplied alpha. An image with premultiplied alpha has its color channels already multiplied by the alpha channel, which means the matte’s edges are already baked into the colors. This is correct for compositing math (it’s what the Over operator expects).
The problem is that premultiplication only works correctly in linear color space. If you premultiply in log, the math is wrong — the color values around the edges are no longer correct, and the comp produces visible halos or fringing.
The standard practice: convert to linear, premultiply in linear, composite in linear, demultiply for any operations that require unpremultiplied color (color correction, for example), then convert back to the deliverable color space at the end.
A compositor receiving plates that were premultiplied in the wrong color space (because someone in the pipeline upstream got the order wrong) has a real problem. The premultiplication can sometimes be undone, but the recovery isn’t free. Catching this in the pipeline test is the only reliable way to prevent it from cascading through the project.
What Goes Wrong
The most common color space mistakes that compositors make:
- Compositing in log. The math doesn’t work; the shot looks “off” in subtle ways that are hard to articulate but easy for an experienced eye to spot.
- Converting log to Rec.709 too early. Once the conversion happens, the highlights and shadows that were in the log file are gone. Working in Rec.709 from the start of the comp throws away most of the camera’s range.
- Mixing color spaces in a single composite. A plate in linear, a CG element in Rec.709, a matte painting in sRGB — all combined in the comp without conversions. The result has color shifts that don’t line up.
- Wrong premultiplication order. Color operations performed before demultiplication, or after re-multiplication. The matte’s edge math goes wrong, and the comp shows halos.
- Inconsistent ODTs. Different shots in the same delivery converted with different output device transforms, producing color shifts between shots that the colorist has to fight.
Each of these is recoverable, in principle. None of them is free to recover in practice. The discipline of color management — converting once at ingest, working in linear, converting once at delivery — prevents most of these from happening.
How FXiation Digitals Handles Color Pipeline
Our compositing pipeline defaults to ACES when the production is set up for it, and to a linear working space with explicit per-shot conversions when ACES isn’t in use. Either way, the math happens in linear, the conversions happen at ingest and delivery, and the spec sheet for every project locks the color pipeline before any work starts.
When we receive plates whose color space isn’t documented, we’ll flag it and ask. Guessing at a plate’s color space is a practice that produces shots that don’t survive a colorist’s review. Asking takes a few minutes and saves hours.
If you’re starting a project and want to confirm your color pipeline is set up correctly before any compositing happens, send us the technical brief. The conversation is short and consistently prevents the most common color management failures from showing up later. Color is one of those areas where the pipeline integration has to be solid before the creative work can move — and the compositors on our side spend their time making shots better, not fighting color space conversions that should have been settled in week one.
Common Questions