Does it require a photograph?

No. Marks can also represent other inputs, such as a vector field, depending on the method.

Can the individual strokes be edited?

Potentially, if the system retains and exposes their descriptions. A flattened image alone does not provide that capability.

Is it the same as neural style transfer?

No. Style transfer describes a different family of techniques; a particular system may combine learned style methods with an explicit stroke representation.

AI Glossary / Computer vision and multimodal AI / Intermediate

Stroke-based rendering

Definition

Stroke-based rendering is a way of creating an image by placing and drawing discrete elements, such as brush marks, ink lines, or dots. A program selects their properties and arrangement to produce a visual effect. The output can resemble a painting, a stippled illustration, or another intentionally stylized representation.

What it means

The image is built from marks that have individual descriptions. A brush mark might have a position, direction, length, width, color, and opacity. Changing those descriptions changes the finished image.

This makes the representation different from simply recoloring every pixel in an existing photograph. There is a collection of marks that can potentially be moved, resized, reordered, or removed. The renderer turns those descriptions into visible pixels or another supported output format.

Stroke-based rendering is a computer graphics approach, not automatically a machine learning system. Some methods follow explicit rules; others optimize a result or use learned components. The presence of a painterly appearance alone does not reveal which approach produced it.

How it works

The designer first chooses a mark model and a goal. For a portrait illustration, the goal might combine recognizable facial features with a limited number of broad strokes. For a flow diagram, marks might need to follow a direction field while remaining readable.

The system selects mark parameters, renders a candidate, and assesses or updates the arrangement. A method may add marks progressively or repeatedly adjust existing ones. The final rendering stage determines how overlapping marks, textures, and transparency appear.

Fidelity and abstraction can pull in different directions. Covering every detail with tiny marks may preserve the source closely while losing the intended simplicity. A useful design therefore specifies which details matter and how much visual simplification is acceptable.

Three practical examples

1. A tourism team turns a licensed landscape photograph into a poster illustration. Broad marks establish the sky and hills; smaller marks preserve a recognizable landmark. A designer checks that the rendering has not changed the landmark's distinctive silhouette.

2. A technical illustrator uses short directional marks to show a simulated flow. The placement should communicate the field without creating dense tangles that readers mistake for physical obstacles.

3. A product designer prepares several stippled packaging concepts. Dot density suggests shading, while important lettering remains a separate, sharply rendered layer. The final printed size determines whether the dots remain distinct.

Think of it this way

Think of composing a picture with pieces from an art kit. Each piece has a shape and a place, and overlapping pieces change what is visible underneath. The analogy explains the importance of individual elements. It does not imply that digital marks behave like real paint: physical pigment mixing, drying, and brush pressure require additional models.

Worked example

Consider one pixel covered by two translucent strokes. Use normalized linear RGB values and ordinary alpha compositing over an opaque white canvas. Each stroke has opacity 0.5.

The white canvas is (1, 1, 1). Applying a red stroke, (1, 0, 0), gives:

0.5 × red + 0.5 × white = (1, 0.5, 0.5).

Applying a blue stroke, (0, 0, 1), next gives:

0.5 × blue + 0.5 × previous = (0.5, 0.25, 0.75).

Reverse the order. Blue over white first gives (0.5, 0.5, 1). Red over that result gives (0.75, 0.25, 0.5).

The same two marks therefore produce different colors where they overlap. Their order is part of the image description, not merely a storage detail. This simplified calculation isolates compositing; it does not simulate the chemical mixing of red and blue pigments or a textured brush surface.

Illustrative case study

Fictional case study: Cedar Trail Tours commissions a series of illustrated walking maps. Its first automated rendering gives every source photograph the same large brush size. The hillside scenes look coherent, but several small bridge entrances disappear into nearby vegetation.

The designer marks those entrances as important regions and permits smaller strokes there. Labels and route lines are composed separately after the illustration is rendered. The team reviews the actual printed map rather than only enlarged images on a monitor.

The revised process treats visual style and navigation accuracy as separate review questions. A pleasant illustration does not establish that a path or crossing is represented correctly. The company keeps the original map data available for that check. This scenario illustrates workflow choices, not a measured commercial result.

Common misconceptions and mistakes

A common mistake is assuming that more strokes always improve the result. More detail can weaken an intended graphic style or obscure the main subject.

Another mistake is treating every painterly image generator as stroke-based. Some systems produce pixels without an editable list of marks.

Finally, do not assume that an appealing still-image method will work unchanged on video. If marks appear at unrelated positions in adjacent frames, the result may flicker even when each frame looks convincing by itself.

Business use

Businesses can use this approach for illustration variants, explanatory graphics, and controlled visual styles. A practical evaluation should include recognizability, editability, output size, rendering time, and consistency across a collection.

Decide which elements require precise reproduction before rendering. Product text, identification labels, and diagram annotations may need separate treatment. Keep the source asset, stroke settings, and approved output together so that later revisions can be reproduced and compared.

Check your understanding

Two translucent marks cover the same area. Can their order be changed without checking the output?

Answer: No. Alpha compositing generally depends on order, so the overlap can change even when every mark retains its own color and opacity.

Related terms

Neural style transfer
Nonphotorealistic rendering
alpha compositing
stippling
vector graphics
procedural generation

Learn next

Learn alpha compositing next to understand overlapping marks. Then compare a flattened image with an editable mark list, and experiment with stroke size while judging both recognizability and visual abstraction.

Sources

Aaron Hertzmann, A Survey of Stroke-Based Rendering, IEEE Computer Graphics and Applications, 2003: https://www.dgp.toronto.edu/~hertzman/sbr02/hertzmann-cga03.pdf

The survey supports the representation and algorithm overview. The compositing calculation, practical examples, and fictional case are original teaching illustrations.

Previous: String kernel Next: Strong connectivity