A 3D model lit normally has a smooth gradient from lit to unlit. A drawing does not; it has a lit shape, a shadow shape, and a line between them. Cel shading is the set of tricks that turns the first thing into the second. Done well, a rendered frame of two fictional adult women can sit beside hand-painted work without announcing which is which.

The ramp is the whole idea

Under the hood, cel shading takes the usual light calculation and pushes it through a lookup: anything above a threshold is fully lit, anything below is shadow. Two bands is the classic look. Three gives you a mid-tone for round forms. More than four and you are back to a gradient, which defeats the purpose.

The threshold position matters more than the number of bands. Put it high and the character reads as mostly shadowed and moody. Put it low and they read as flat and bright. Most anime-styled work sets it so roughly two thirds of the form stays lit, which keeps faces open and keeps the shadow shape doing real work as a design element.

Outlines do half the job

There are two common ways to get a line. Inverted-hull pushes a copy of the mesh outward and flips its normals, giving an outline whose thickness you control per object. Screen-space methods find edges from depth and normal buffers, which catches interior detail the hull method misses. Many productions run both and mask where each applies.

A uniform line thickness looks mechanical. Real drawings vary weight: heavier where forms overlap or turn away, lighter on interior detail. Feeding a thickness map into the hull, or scaling line width by distance, gets you most of the way there. This is the single change that moves a render from "3D pretending" to something a reader accepts.

Faces need their own rules

A face is where the illusion is tested, because viewers know faces. Ordinary lighting drops a shadow under the nose and across one eye socket, which in a drawn style reads as dirt. The common fix is a separate light rig for the head, plus a custom normal map that smooths the face into a near-sphere so shading bands sweep cleanly across it.

Hair gets the same treatment for the same reason. Real hair geometry produces noisy shadow shapes; drawn hair produces a few clean masses with a highlight band. Simplifying hair normals into large soft groups is standard practice, and it is why cel-shaded hair often has that one glossy stripe rather than a scatter of small speculars.

Where the illusion falls apart

Grazing angles are the first crack. When a surface turns almost edge-on to the light, the band boundary crawls and flickers between frames. Dense geometry is the second: a highly detailed mesh generates band edges that a human would never draw. The answer to both is simplification, either in the mesh or in the normals feeding the shader.

Motion is the third. A drawn frame can cheat perspective and anatomy freely; a rendered one cannot, because the model stays consistent from every angle. That consistency is a strength for continuity and a weakness for expression, and it is why hybrid pipelines exist at all.

Making a cel-shaded model read as a drawing

Before any of that, decide what the style is copying. Cel shading is not one look; a nineties television palette, a modern web comic, and a painted feature all use flat bands differently. Pick a single reference frame and match it, because a shader tuned against a vague memory of anime ends up looking like every other default, which is the outcome nobody wants after all that work.

Work in this order: set the ramp, fix the normals, then add lines, then add the small effects. Doing it backwards means tuning outlines against shading you are about to change. Keep a hand-drawn reference frame open beside the viewport and compare shadow shapes, not colours, because shape is what the eye is actually matching.

Colour choices sit on top of all this, and they carry their own meaning; our notes on building a palette cover that side. If you want the pipeline argument rather than the shader one, read the hybrid pipeline piece, and the studio route for making an original scene is on its own page.