Can Claude generate images?

The short answer

Claude reads images very well: screenshots, charts, a photo of a whiteboard. It can also draw anything made of code, like SVG icons, diagrams and charts.

What it can't do is turn a prompt into a picture. There is no image model inside Claude, and Anthropic's help center says so.

Why Claude has no image generator

Anthropic focuses Claude on text, code and agents. Instead of a built-in generator, Claude calls outside tools. For images that's a plus: it can pick the right model, look at the result and try again.

Add image generation in one command

MCP is the open standard that lets Claude use outside tools. The Imference MCP server gives it 75 image and video models. The server is free and open source; you pay per image.

Claude Code

claude mcp add imference -e IMFERENCE_API_KEY=your-key -- npx -y imference-mcp

Claude Desktop, Cursor and other clients

Add this to the MCP config (Claude Desktop: Settings, Developer, Edit Config. Cursor: Settings, MCP), then restart. Needs Node.js 18 or newer.

{
  "mcpServers": {
    "imference": {
      "command": "npx",
      "args": ["-y", "imference-mcp"],
      "env": { "IMFERENCE_API_KEY": "your-key" }
    }
  }
}

Get a key on the pricing page, then ask: "Make a banner of a lighthouse in a storm. Pick the best model."

What Claude does with it

1. Browses the catalog with list_models: style, price, prompt format.
2. Reads the model's card with get_model, so the prompt fits that model.
3. Generates in the right format, banner, portrait or square.
4. Looks at the image and tries again if it missed.

That last step is what makes the difference: Claude sees what it made.

Which models can it use?

74 image models and 1 video models, from $0.003 an image:

AnimaSDXLIllustriousKrea 2PonyZ-ImageOpenAIMinimax

Ask for "the cheapest photoreal model" or "an anime model" and Claude finds it. Browse them all.

Can Claude generate images for free?

• Through Claude: from $0.003 per image, no subscription. $5 buys thousands of images at that price.
• Free: Imference Desktop runs models on your own GPU at no cost. Claude doesn't drive it, but the images are free.
• No account: give the server a wallet instead of a key and it pays each image in USDC with x402, under a cap you set.

Can Claude edit images or make videos?

Edits: some models accept a source image. Claude checks the model card, then applies your change.

Videos: yes, with Wan 2.2 and MiniMax H3 (with sound). They take a few minutes; Claude waits and shows the file.

Claude vs ChatGPT for images

ChatGPT has image generation built in, which is simpler for a one-off picture. Claude with MCP wins once you generate often: dozens of specialized models instead of one, prompts written for each model, pay per image, and images that land straight in your project.