Can Claude generate images? What Claude can and cannot do with images (2026)

No: Claude reads images but does not create them. Anthropic's documentation says Claude cannot generate, edit or create images. On claude.ai, Claude can draw diagrams, charts and interactive visuals as HTML and SVG code, and Anthropic's Claude Design produces designs, prototypes and slides, but Anthropic does not describe either as an image generator. To generate images from code, call an image model such as Nano Banana, GPT-Image-2 or Ideogram v3 with the same Kunavo key; the four in the table below start at $0.02625 per image.

Last reviewed on October 22, 2026 against Anthropic's own pages, linked below.

What Claude's API returns

All current Claude models accept text and image input and return text output, according to Anthropic's models overview. The vision guide answers the question directly: Claude is an image understanding model only, and it cannot generate, produce, edit, manipulate or create images. A Messages API response is therefore text, tool calls included, never an image file.

Claude can still write code that draws: an SVG, an HTML canvas, a chart script. When that code runs, it produces a picture, but what the model returned was the code.

What claude.ai does with images

Anthropic's help center (updated March 16, 2026) says Claude does not generate photos or illustrations the way image-generation tools do. In a conversation it can build diagrams, charts and interactive visuals with HTML and SVG, a feature in beta on Claude web and desktop, and it can view and analyze images you upload.

Claude Design, which Anthropic Labs introduced on April 17, 2026, creates designs, prototypes, slides and one-pagers with Claude for Pro, Max, Team and Enterprise subscribers. Anthropic's announcement does not describe it as an image generator, and it is a claude.ai product, not an API model.

Generate an image with the same key

On Kunavo, the key that calls Claude also calls image models, through the OpenAI-compatible POST /v1/images/generations:

curl https://api.kunavo.com/v1/images/generations \
  -H "Authorization: Bearer $KUNAVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nano-banana",
    "prompt": "a red origami crane on a white desk, soft window light",
    "size": "1024x1024"
  }'
# The image URL is in data[0].url of the response.

Swap model for any image model on the key; parameters, editing and asynchronous jobs are in the image docs.

Price per image, by model

A price ending in + is the lowest tier: higher resolutions, or for Ideogram the Balanced and Quality renders, cost more. The tier-by-tier comparison is in the image API comparison, and two models have pages of their own: Nano Banana API and GPT Image 2 API.

From Claude Code

Claude Code runs on Claude models, so it writes text and code, not images. A session can still produce one by calling the images endpoint with a Kunavo key; the Claude Code setup and the image docs cover both halves.

Frequently asked questions