Can Claude generate images?
What Claude produces (text, code, SVG, HTML/CSS), what it does not (photos, raster illustrations), and how to use an image generator with the catalog's resources.
Updated on
No, Claude does not generate images the way an image generator does: no photos, no realistic illustrations. Anthropic's documentation is blunt about it: Claude is an image understanding model, and it analyzes the images you send. What it writes very well is text and code, and some of that code draws.
What Claude can do
Anthropic's help center states that Claude can build diagrams, charts and interactive visuals with HTML and SVG. In a coding agent the principle is the same: it writes a file, you open it.
- SVG: simple logos, icons, pictograms, diagrams, patterns. The output is vector, light and editable by hand.
- HTML and CSS: mockups, banners, business cards, social media visuals you can export to PNG from the browser afterwards.
- Charts in code: line charts and bar charts built from your data with a charting library.
- Prompts for an image generator: Claude writes, you or a tool runs.
- Image analysis: it reads a screenshot, a logo or a mockup and turns it into a description or code.
What it does not do
It does not produce raster images (photos, 3D renders, digital paintings) and does not retouch an existing image. For that you need an image generator: a separate service with its own account and terms.
Pick the right route
| Need | Route | Result |
|---|---|---|
| Icon, simple logo, diagram | Claude writes an SVG | Vector file |
| Banner, page visual | Claude writes HTML and CSS | Page to export as an image |
| Numeric data | Claude writes a chart in code | Chart faithful to the data |
| Photo, illustration, render | Claude writes the prompt, an image generator makes the image | Raster image |
Using an image generator
Two ways, your choice.
- Prompt copied by hand. Ask Claude for a precise prompt (subject, style, framing, lighting, format), paste it into the generator you prefer, then come back with the result to iterate.
- Tool connected through MCP. The Model Context Protocol lets Claude Code use tools beyond its built-in set. If an image generator publishes an MCP server, you can connect it with
claude mcp add: the agent then calls the generator and gets the image back. Check the generator's own documentation: its server, account and billing depend on neither Anthropic nor PalateKit. See also the article on skills and MCP.
With the catalog's resources
- The catalog's prompts cover text, image and video: take one as a starting point and have Claude adapt it to your subject. The how-to is in the article on image and video prompts.
- DESIGN.md files give the agent your colors, typefaces and rules, so the SVG or HTML it writes follows a brand guide.
- Skills add reusable methods; install them in your agent, for example Claude Code, as explained in the installation article.
- To animate the result, the motion templates take over.
Example requests
For an SVG:
Draw a magnifying glass icon as an SVG, 64 x 64, 4 px stroke, color #1F2937, transparent background. Save it to public/icons/search.svg.
For an image generator:
Write a prompt for an image generator: product photo of a white ceramic mug on an oak table, side window light, blurred background, 4:5 format. Give a short version and a detailed version.
Always check the SVG in a browser before using it: a drawing described in code is verified by eye, not by reading.
Sources
Frequently asked questions
Can Claude create a photo or an illustration?
No. According to Anthropic's documentation, Claude understands and analyzes images, but it does not generate photos or illustrations the way image generators do.
What visuals can it produce, then?
Code that renders: an SVG file, an HTML and CSS page, a chart, a diagram. You open them in a browser or export them as images.
How do I get a real image with my agent?
Ask Claude to write the prompt, then send it to an image generator. If the generator offers an MCP server, the agent can also call it itself, provided you have connected it.
