Compositor
Combines plates, CG renders and paintings into one image that reads as photographed.
- Compositing artist
- Comp artist
- Finishing artist
- 2D artist
- Visual effects artist
What they do
The day starts with a plate and a stack of renders that have to become one photograph. First the extraction: keying green or blue screen, refining with roto, patching from prep and clean plates, despilling, and holding edges that carry motion blur and hair. Then integration, which is mostly light. Black level and highlight rolloff get matched, atmospheric haze goes in between camera and subject, CG is relit from its render passes, and everything the taking lens did is put back: distortion, chromatic aberration, bloom, halation, grain, and the particular softness of that stop. The work happens in a node graph in Nuke, on scene-linear float data usually carried in OpenEXR, inside an OpenColorIO-managed pipeline so the viewer transform agrees with what the grade will do. Render passes and, increasingly, deep data are the raw material; deep files store colour and opacity through depth for every pixel, which is how elements are combined without re-rendering holdout mattes each time animation changes. There is also the tracking and matchmove that has to be exact, roto nobody will ever see, and a final ten per cent that takes as long as the rest. Days are versions: a comp goes to dailies, comes back with a note about one edge, and goes round again.
What they decide
The calls that belong to this role and no other.
- How each element is extracted and where its edge sits
- How the shot is integrated: light matching, atmosphere, lens artefacts and grain
- The structure of the node graph, and therefore what can be changed cheaply after a note
- Whether delivered CG is usable or has to go back to lighting
- What is fixed in comp and what gets escalated when a plate has a real problem
Principles of the craft
- Match the plate's flaws, not its ideal: the plate has grain, aberration, a soft corner and a slightly wrong black. An element cleaner than the plate reads as fake, so degrade the new thing toward the old one rather than improving the old one.
- Integrate through light and lens, not through edges: beginners fight the matte, but comps usually fail on wrong black level, missing atmosphere or absent lens behaviour. A slightly imperfect edge in a correctly lit shot survives; a perfect edge in a wrongly lit one does not.
- Work scene-referred and linear: add, multiply and blur in linear float data, because that is how light actually combines. Compositing in a display-referred space is why highlights turn grey and edges go dark.
- Hold the black levels: lifted or crushed blacks between elements are the fastest tell in an image, and they are the first thing to check on any shot that reads wrong for no obvious reason.
- Build the tree so somebody else can change it: comps take notes for months. Named nodes, separated concerns and an exposed control for the thing that will certainly be asked about turn a note into a ten-minute change.
- Nothing in the real world is a hard cut: there is always atmosphere, bounce, spill and interaction between an object and its surroundings. Putting that back is what makes an element occupy the space rather than sit on top of it.
Common mistakes
- Chasing a perfect key on a badly lit green screen instead of combining a soft key with roto and paint
- Working with the wrong viewer transform, or in a display-referred space, then being surprised when the grade breaks the shot
- Adding grain last as a blanket over the whole frame at a strength that does not match the plate per channel
- Cranking contrast and saturation on the element so that it stands out
- A node tree only its author can navigate, so every note becomes a rebuild
- The competent compositor's mistake: a technically immaculate shot that ignores where the eye goes, so the element is integrated but not staged
How to learn it
Exercises and routes in, not a reading list.
- Composite a still CG object into a photograph you took yourself, using the light direction you actually measured on location. Then do it again from memory with no reference and compare the two.
- Key one green screen shot three ways: keyer only, keyer plus roto plus despill, and roto only. Time each and note which held up at the frame edges and in motion blur.
- Take a comp you are happy with and view it through two different display transforms. Fix whichever one breaks, and write down what you had been relying on without knowing it.
- Match grain properly: sample the plate's grain per channel, degrain, comp, then re-apply your sampled grain. Compare that against simply adding a preset.
- Comp the same element twice, once matching black level and atmosphere before anything else, once matching colour and saturation first. Note which route got you to a believable shot faster.
Machine-learning tools are already ordinary in comp for roto, matte extraction, cleanup, depth estimation and upres, and they shorten the parts of the job that were never the interesting parts. Generated elements arrive already baked, with their own look and grain and no render passes, so integrating them is harder than integrating CG you controlled, and colour space has to be handled explicitly because it is rarely stated. Diagnosing why a shot does not read, and knowing the answer is usually black level and atmosphere rather than the edge, is the part that does not automate.
Tools and kit
- Nuke, with Blackmagic Fusion, Flame and After Effects on some jobs
- OpenEXR scene-linear multi-channel files, including deep data
- OpenColorIO configurations and ACES colour management
- Render passes and AOVs from lighting
- Keyers, roto, paint and despill tools
- 2D and 3D tracking, matchmove and camera solves
- Grain, lens distortion and optical artefact tools
- Versioning and review systems for dailies
Works with
- vfx-supervisor
- matte-painter
- lighting-td
- fx-td
- animator
- colorist
- editor
- assistant-editor
The models that touch this work
The parts of the directory relevant to visual effects, rather than all of it.
The vocabulary, on the other side
Lenses, film stocks, lighting patterns and mood recipes, written in the words this department already uses. Useful if you are briefing a model rather than a crew.
Also in Visual Effects
CG Generalist
Takes a shot from model to render single-handed, covering every 3D step in between.
Visual Effects
Matte Painter
Paints and builds the parts of a location that were never there, so they read as photographed.
Visual Effects
Motion Graphics Designer
Designs and animates type, graphics and screen content so information moves legibly.
Visual Effects