Claude canvas-design skill

Claude does not paint pixels the way an image model does, but it can draw with code. Anthropic's canvas-design skill gives that drawing a method: decide on a visual idea first, then build the image from it. Its description says to use it when you ask for a poster, a piece of art, a design or another static piece. It is an open-source example skill under the Apache 2.0 license. For the basics of skills, see how Claude Skills work.

What the skill does

The SKILL.md splits the work in two:

1. A design philosophy, saved as a markdown file. Claude names an aesthetic movement in one or two words and describes, in four to six paragraphs, how it uses space, color, scale, rhythm and hierarchy.
2. The canvas. Claude expresses that philosophy as a single-page PDF or PNG, mostly visual with very little text.

The rules are strict about finish. Text stays sparse, fonts come from a bundled font folder, and nothing may overlap or run off the page. After the first version, Claude takes a second pass to refine what is already there instead of adding more elements. You can ask for more pages, which it bundles into one PDF or a set of PNGs. The skill also tells Claude to make original work and never copy an existing artist.

When to use it

• An event poster or cover image with a clear visual concept.
• Abstract art for a slide background, a report cover or a print.
• Exploring a visual direction you can describe in words.

For photos, illustrations and realistic scenes, an image model does better. See our guide to AI image generators and our AI image prompts. For moving, code-based art, use algorithmic-art; for web interfaces, frontend-design.

How to install it

Claude.ai: the anthropics/skills README says its example skills are already available on paid plans. Turn on code execution and check Customize > Skills.

Claude Code:

/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills

Example request

Use the canvas-design skill to make a poster for [event or idea], [size or orientation]. The only text should be "[title]" and "[date or one short line]". Mood: [two or three words]. Save it as a PNG.

Limits

• It draws with code, so it suits geometric, typographic and abstract work. For photographic images, an image model is the better tool.
• Text is kept to a minimum by design, so it is a poor fit for infographics or flyers full of details.
• Anthropic labels its example skills as demonstrations and advises testing before you rely on them.

FAQ

Is canvas-design the same as Canva?

No. It is a skill that has Claude draw an image in code and save it as PNG or PDF. It does not connect to Canva.