September 20, 2026
Claude does not generate images itself: add a generator to Claude Code in 2 commands
Claude only understands images — that is what Anthropic's docs say. Image generation can be added to Claude Code with a plugin in two commands.
Ethan Mollick
@emollick
Claude has no image generator, and that is a gap in agent projects centered on knowledge work. It can draw and model objects in code quite well, but image generators from Google and OpenAI give their AI far more options for presentations, mockups, infographics, and more.
· 51.5K views
An agent is putting together a presentation and hits a wall at the very first slide: there is nowhere to get an image.
Anthropic does not hide this. The FAQ on its Vision page states it plainly: Claude only understands images and cannot generate, create, edit, or alter them. Google and OpenAI have generators in the same stack, so their agents can assemble slides, mockups, and infographics on their own.
Anthropic addressed its part of the task from another angle. Claude Design launched on 17.04.2026 as a research preview on Opus 4.7 and creates slides, one-pagers, wireframes, mockups, and pitch decks with exports to HTML, PDF, PPTX, and Canva. It is included in Pro, Max, Team, and Enterprise, counts against subscription limits, and is off by default for Enterprise until enabled by an admin.
A two-command plugin. Claude Code has a ready-made image-generator: `/plugin marketplace add dair-ai/dair-academy-plugins`, followed by `/plugin install image-generator@dair-academy-plugins`. The third step is adding a `GEMINI_API_KEY` from Google AI Studio. Under the hood, it runs Nano Banana Pro, and the agent loops through the image itself, refining it against the stated criteria (DAIR.AI Academy blog, 03.02.2026).
A second plugin lets you choose a provider: `/plugin marketplace add hex/claude-marketplace` and `/plugin install claude-image-generation`. It uses gemini-3-pro-image up to 4K, gpt-image-2 up to 1536×1024, grok-imagine-image-2.0 up to 2K, and OpenRouter. Its commands are `/generate-image <prompt>` and `/generate-image --edit ./photo.png <instruction>`, and a key from any one provider is enough.
A third path works without plugins at all. The line `npx nano-banana-mcp` in the mcpServers block with the GEMINI_API_KEY variable connects image generation and editing in both Claude Code and Cursor.
The price. Google has no free generation tier: Nano Banana Pro costs $0,134 per 1K or 2K image and $0,24 per 4K image, while Nano Banana 2 costs $0,067 per 1K image (Gemini API docs, 20.09.2026).
Claude Design has no image generation either: it lays out slides, while the images for them still have to come from elsewhere.
Source
