Look development artist

What they do

Look development takes a modelled, UV'd asset and decides what it is made of. The artist builds shading networks in a renderer's material system, paints or generates texture sets in Substance Painter or Mari across UDIM tiles, and produces the maps a physically based shader needs: base colour, roughness, normal or displacement, transmission, subsurface. Then they test the asset against a standard set of lighting environments, a grey and chrome ball, and a turntable, because the deliverable is not a pretty frame but an asset that holds up under any light the show will throw at it. Iteration is the working day, and iteration speed is the constraint the whole discipline is organised around, which is why renderers have grown interactive and hybrid CPU-GPU modes specifically for this loop. Hair, fur, skin and eyes are separate specialisms; art-directable hair shading models exist because the physically correct version was not controllable enough for a director's notes. Published work on look development at Pixar and Disney Animation is largely about that tension between physical correctness and artistic control. The unglamorous parts are variant management, matching an asset to concept art under three different lighting keys, and being the person who discovers the UVs are wrong.

What they decide

The calls that belong to this role and no other.

• The material response of each surface: how rough, how transmissive, how much subsurface
• How a shading network is structured, and which parameters are exposed for lighting and shots to override
• What is painted by hand, what is procedural, and what is a scan or photographic source
• Texture resolution and UDIM tile allocation per asset, against memory budgets
• Which lighting environments an asset must be approved under, with the CG supervisor
• Whether to hold physical correctness or break it for art direction, recommended to the director

Principles of the craft

• Look is a response, not an appearance: you are not authoring a picture, you are authoring how a surface behaves when light hits it. Judge every material under at least three environments, including one that is unflattering, because the shot that breaks your asset will be the one lit from behind.
• Roughness carries the read: most of what makes a surface identifiable as wet stone, worn leather or new plastic lives in the variation in specular roughness, not in the colour map. Paint roughness with as much intent as albedo and the material stops looking like a texture on a shape.
• Physical correctness is a starting point you are allowed to leave: production shading models get art-direction hooks added precisely because a director's note about a character's hair cannot be answered by a more accurate model. Know the physical baseline well enough to know what you are breaking and by how much.
• Iteration speed is a creative variable: look development is a converging loop of small judgements, and a five-minute test render gets you three tries an hour while an interactive one gets you thirty. Setting up cheap look tests, low-resolution, few samples, one asset, is part of the craft.
• Build for the shot, expose the right handles: lighting will need to push the eye brightness or knock back a costume in one sequence. Decide in advance which parameters they can safely override, and hide the ones that would break the asset's identity.

Common mistakes

• Approving a material in one flattering HDRI, then discovering it is plastic in daylight and dead in a night interior.
• Everything at the same mid roughness, which produces a whole film of surfaces that read as the same substance painted different colours.
• Texture resolution chosen by habit rather than by shot: 8K tiles on a background prop that never comes closer than a wide, blowing the memory budget.
• Painting light and shadow into the base colour map, which then fights every light in every shot and cannot be undone downstream.
• The experienced artist's mistake: a physically exquisite asset that no lighting artist can push, because every useful parameter is buried inside a locked network.

How to learn it

Exercises and routes in, not a reading list.

• Build one material in Blender's Cycles and render it under five HDRIs, including a hard single-source and a flat overcast. Do not adjust between them. Wherever it fails, fix the material rather than the light.
• Take one object, a chipped enamel mug, and reproduce it from a photograph you shot yourself with a grey card in frame. Match it side by side under the same light direction. The gap in the specular is where the learning is.
• Paint the same asset twice: once with a detailed colour map and constant roughness, once with a flat colour and a fully painted roughness map. Render both. Note which one reads as a real material.
• Download the Disney Animation Moana Island Scene data set and read how production materials are actually specified and instanced across thousands of assets, then rebuild one of its material ideas from scratch in your own renderer.
• Set yourself a test-render budget: no look test longer than 60 seconds. Build the low-sample, low-resolution setup that makes that possible, and count how many more variations you try in an evening.