Character modelling
Character modelling is the discipline of building the three dimensional geometry, surface detail and material definition of a figure, whether human, creature or stylised, so that it can be rigged, animated and rendered in a game engine or a film pipeline. It sits between concept art and the rigging, animation and lighting that follow, and is judged on two things at once: whether the character reads well as an image, and whether it behaves well as a technical asset.
In larger productions the work is split further into character design, sculpting, look development, grooming and technical art, but the core sequence is consistent across studios and media.
Common tools are ZBrush for sculpting, Maya for modelling and retopology, Substance 3D Painter and Designer for texturing, Unreal Engine for real time work, and Keyshot or Photoshop for presentation.
What a character artist does
A character artist translates a two dimensional design into a production ready three dimensional asset. The interpretive side of the role means reading a concept sheet, understanding the intent behind proportions and costume, and resolving what the concept does not show, such as the back of a garment or how a strap fastens. The technical side means meeting a specification: polygon budgets, texture resolution, naming conventions and engine import rules.
Character work overlaps with, but differs from, environment art. Environments are modular, reused and optimised for coverage. Characters are singular, seen close to camera, and above all they move, which makes deformation the dominant constraint. Many artists work across both. In the case of Tomasz Strzalkowski, a Warsaw based artist associated with the studio People Can Fly, character modelling and character design appear among the listed skills alongside environment work, digital sculpting, texturing and lighting. His credited work includes a King Kordax concept for Aquaman and the Lost Kingdom, alongside game credits including Painkiller, additional level design on Gears of War (2006) and lead environment artist on Bulletstorm (2011).
From concept and reference to high resolution sculpt
Reference and blockout
Before any geometry exists, the artist assembles reference: anatomical plates, photographs of real garments and armour, and material samples. Reference is not copied, it keeps invented forms plausible, determining how seams fall and how leather creases differently from cotton or metal.
The blockout that follows is a rough statement of proportion and silhouette, deliberately crude so that large errors are cheap to fix. Proportion mistakes made here become expensive later, because sculpting detail onto wrong forms wastes the detail. Blockouts are checked against a reference height and the surrounding scale.
High resolution sculpting
Digital sculpting tools allow the artist to push millions of polygons as if working clay. Sculpting proceeds from primary forms, meaning the large masses and their gesture, to secondary forms such as muscle groups and major fabric folds, and finally to tertiary detail such as pores, stitching and wear. This order preserves readability, since detail applied before the primary forms are correct tends to flatten the sculpt.
Anatomy, silhouette and readability
Anatomy knowledge underpins believable characters even when the character is not human. Understanding skeletal landmarks, muscle origins and insertions, and how skin drapes over structure allows the artist to make deliberate deviations rather than accidental ones. Stylised characters are not exempt: exaggeration reads as intentional only when the underlying structure is understood.
Silhouette is the character outline seen as a solid shape. In games it is often the first thing a player perceives, carrying identity and threat level at distance. Artists commonly test a character by viewing it as a flat black shape. Strong silhouettes combine a few large contrasting masses rather than uniform surface noise.
Retopology, UVs and baking
Retopology and edge flow
The sculpt is not directly usable in production. Retopology rebuilds the surface as an efficient, animation friendly mesh. The critical concept is edge flow: the direction in which loops of polygons run. Edge loops should follow the anatomical lines of deformation, meaning concentric loops around the eyes and mouth, loops following the brow and nasolabial fold, and enough rows of geometry at elbows, knees, shoulders and fingers.
Topology decides whether a face and joints deform well. A mesh with loops running across a joint rather than around it will pinch and collapse when bent, and a face without proper loops around the mouth cannot form clean vowel shapes or asymmetric expressions, no matter how good the rig is. Quad based topology is preferred because it subdivides predictably; triangles and n gons are tolerated only in rigid, non deforming areas such as armour plates.
UV layout
UV unwrapping flattens the surface into two dimensional coordinates so textures can be applied. Good UV layout minimises stretching, hides seams behind costume edges or hair, allocates resolution proportionally so the face receives more pixels than the boots, and packs shells efficiently. Characters are often split across multiple UV tiles or texture sets.
Baking
Baking transfers information from the high resolution sculpt to the low resolution mesh as texture maps. The most important is the normal map, which encodes surface direction so the low polygon mesh catches light as though it carried the sculpted detail. Ambient occlusion, curvature and thickness maps are baked alongside it and later drive procedural wear, edge highlights and subsurface effects.
PBR texturing and materials
Physically based rendering, or PBR, describes surfaces using parameters that correspond to measurable physical properties, so an asset behaves consistently under any lighting. The metalness workflow uses base colour for the diffuse albedo with lighting information removed, roughness for how scattered the reflection is, metalness as a near binary mask separating metals from dielectrics, and the baked normal map for fine surface direction. Different materials demand different treatments:
- Skin relies on subsurface scattering, because light penetrates and diffuses beneath the surface. Roughness varies across the face, the forehead and nose being oilier than the cheeks.
- Cloth is defined by weave scale, fibre sheen and how dirt and wear accumulate at seams and cuffs.
- Hair is built as layered cards with alpha transparency in real time work, or as groomed strands for film, and depends on anisotropic highlights running along the strand direction.
- Armour and metal use metalness with roughness variation to tell a history: polished where handled, scratched at exposed edges, oxidised in recesses.
Handoff to rigging and animation
The modelled character is passed to rigging, where a skeleton is built and the mesh bound to it through skinning weights. Blend shapes are added for facial expression and for fixing volume loss at extreme joint angles. This handoff is where earlier topology decisions are proven or exposed. Missing edge loops cannot be invented by a rigger, and a face without loop structure limits the whole facial animation system built on it. Character artists therefore test their own meshes in extreme poses before delivery, checking for pinching, intersection and volume collapse.
Real time budgets, LODs and the film comparison
Real time characters live inside a frame budget. Polygon counts, texture memory, material and bone counts and shader complexity are all constrained, and the limits differ between a player character, a crowd enemy and a background figure. Levels of detail, or LODs, are progressively simplified versions of the mesh swapped in as the character recedes from camera, reducing polygon count and often simplifying shaders and bone influences. Poorly authored LODs show up as visible popping at the swap.
Film and cinematic characters relax most of these limits. Geometry can be subdivided at render time to millions of polygons, hair simulated as individual strands, cloth run through physical simulation rather than skinning, and textures spread across many high resolution tiles. Rendering is offline, so a frame may take hours, whereas a real time character must arrive in milliseconds. The judgement about anatomy, silhouette and material behaviour is shared, but the technical answers diverge sharply.