If you're wondering whether you can create images with Claude, the answer is a little more nuanced than you might think. Here's the spoiler: while Claude can't generate images on its own, you can still use Claude models to create beautiful AI images with the right tools and workflow. In this article, we'll break down what Claude can and can't do when it comes to working with images, and show you how to use it as part of an AI image creation pipeline.
Claude doesn't have a built-in image generation model, and Anthropic doesn't offer its own image generation model either. This means you can't generate images directly in Claude unless you use a third-party integration.
So, what happens when you ask Claude to create an image? Instead of generating one, Claude will typically explain that it can't create images directly and offer to write a prompt you can use with an image generator instead.
While Claude cannot generate images, it can:
Claude is good at understanding images. You can upload a screenshot and ask it to pull out the text, or share a chart and ask what it shows. On claude.ai, you can attach up to 20 images to one message.
Claude can build simple visuals with code. When you ask for a chart or a diagram, Claude writes it in SVG or HTML, and the result appears right in the reply. Anthropic calls this feature custom visuals. It's still in beta and works on the web and in the desktop app.
This is a good way to get a flowchart, a timeline or a simple icon. But Claude can't make a photo or a detailed illustration this way, since everything it draws is built from basic shapes.
Claude can also create files in the chat. For example, you can ask it to make a presentation or save a chart as a PNG image. This works on all plans, including the free one.
For mockups, slides and one-pagers, Anthropic has a separate tool called Claude Design. It doesn't have an image model either, and Anthropic says it "isn't built for image generation, so it's a poor fit for logo design."
Connectors let Claude use other services. With the Canva connector, Claude can generate new Canva designs right from the chat. The Adobe connector gives Claude access to Photoshop, Firefly, Express and other Creative Cloud apps, so it can edit your photos.
In both cases, the image is made by Canva or Adobe tools, and you need an account in that service.
When you ask Claude to draw something, it writes SVG code, a list of shapes with coordinates and colors. To see how this compares to a real image model, we gave the same lighthouse prompt to Claude Opus 5.5 and GPT Image 2.5 Sunburst.
Claude's lighthouse is made of rectangles, triangles and gradients, while GPT Image 2.5 produced a photorealistic picture.
Yes. ChatGPT and Gemini have their own image models, so they can create images right in the chat. ChatGPT uses GPT Image 2.5, and Gemini uses Nano Banana 2.
| Generates images | Image model | Understands images | |
|---|---|---|---|
| Claude | No | — | Yes |
| ChatGPT | Yes | GPT Image 2.5 | Yes |
| Gemini | Yes | Nano Banana 2 | Yes |
Many people still prefer Claude over ChatGPT or Gemini for writing and coding, so they need a separate tool for images.
If you work in Claude and need images, you have three options:
We'll show the third option using our own app, Overchat AI.
Overchat AI has Claude and image models in one app, so you can write a prompt and generate an image without switching tabs. As an example, we'll turn a summer photo of a city street into a winter evening.
While you can't ask Claude to create a new image directly, you can use it to create a professional image prompt. For example, we asked Claude Opus 5.5 to write a prompt that changes the setting of an attached image to winter.
Claude first described what's in the photo and then rewrote the scene for winter, with snow on the ledges and frozen vines on the buildings.
Copy the prompt from Claude's reply to use it in the next step.
Overchat AI has 15 image models, including Nano Banana 2, GPT Image 2.5, Seedream 5 and Flux 2 Max. Open the image generator and choose a model in the Preferences panel. For this example, we chose Nano Banana 2.
To keep the same street in the result, attach the original photo under Add References. The model will use it as a base and change only what the prompt describes. Then paste Claude's prompt into the text field.
Click Generate to get the image. Nano Banana 2 kept the street and the buildings from the original photo and turned the scene into a winter evening.
If something in the result is off, you can ask Claude to change the prompt, for example to make the sky darker, and generate the image again.
Superchat is a mode in Overchat AI where an AI agent picks the image model for you. You describe the image in the chat, and before generating, the agent shows which model it chose and how much the image will cost.
We sent Superchat the same lighthouse prompt. The agent chose GPT Image 2.5 Sunburst and asked us to approve 25 credits.
The image was ready 25 seconds after we approved it, and the agent offered a few quick edits underneath.
In Canva or Adobe
Canva and Adobe connect to Claude for free, but to create designs or edit photos through them, you need a plan in that service:
| Canva plan | Price | AI uses |
|---|---|---|
| Free | $0 | up to 20 |
| Pro | $144 per year | up to 200 Premium or 20 Ultra |
| Business | $250 per year per person | up to 400 Premium or 40 Ultra |
Canva also has monthly billing, but the yearly plans are at least 16% cheaper.
| Adobe Firefly plan | Price per month | Generative credits per month |
|---|---|---|
| Free | $0 | a few daily generations |
| Standard | $9.99 | 2,000 |
| Pro | $19.99 | 4,000 |
| Pro Plus | $49.99 | 10,000 |
| Premium | $199.99 | 50,000 |
All paid Firefly plans include unlimited standard image generations, and the credits go to premium features such as video and partner models.
Through an image model API
In Claude Desktop and Claude Code, you can connect an image model through MCP and pay the provider only for the images you generate. Google's prices for Nano Banana models in the Gemini API depend on the resolution:
| Resolution | Nano Banana 2 | Nano Banana Pro |
|---|---|---|
| 0.5K | $0.045 | — |
| 1K | $0.067 | $0.134 |
| 2K | $0.101 | $0.134 |
| 4K | $0.151 | $0.24 |
OpenAI charges for GPT Image 2.5 by tokens, so the price of each image depends on its size and quality.
In Overchat AI
Overchat AI has Claude and image models in one app, and both use credits from the same subscription:
| Overchat AI plan | Price | Credits |
|---|---|---|
| Pro weekly | $5.99 per week | 2,000 per week |
| Pro monthly | $14.99 per month | 8,000 per month |
| Pro yearly | $59.99 per year | 8,000 per month |
If you run out of credits, you can buy an extra pack: 4,000 credits for $9.99, 11,000 for $24.99 or 50,000 for $99.99. On the monthly plan, the requests from this article cost:
| Request | Credits | Price |
|---|---|---|
| Claude Opus 5.5 prompt from the first step | 110 | about $0.21 |
| Nano Banana 2 image at 1K | 100 | about $0.19 |
| GPT Image 2.5 Sunburst image at medium quality | 25 | about $0.05 |
In total, the winter street example cost 210 credits, or about $0.39. On the yearly plan, credits are cheaper, and a Nano Banana 2 image costs about $0.06.
Claude is a language model, so it can read text and images but can only reply with text. As of this writing, Anthropic hasn't added an image model to Claude.
Anthropic hasn't announced plans to add image generation. Its newer tools, Claude Design and Claude Motion, also work without image or video models.
No. Claude can tell you what to change in a photo, but it can't edit the file. To edit a photo, you can upload it to an image-to-image model with a prompt from Claude, as we did here, for example, or use the Adobe connector.
Yes, Claude can draw a simple logo in SVG, although the results can be hit or miss.
Yes, if you add another tool to the pipeline. On the free plan, Claude can write an image prompt that you then paste into a free image generator.
No. Claude Pro and Max give you higher limits and more models, but none of the plans include an image model.
