Claude can't make images in the sense most people mean. Anthropic's help center says plainly that Claude doesn't create photos or illustrations, unlike image-generation tools. Claude does work with pictures in other ways: it builds diagrams and charts in HTML and SVG, saves charts as PNG files, reads images you upload, writes prompts for image generators, and drives Adobe and Canva through connectors. Here is what each one gets you, and where to go when you need an actual picture.
Can Claude make images now, in September 2026?
As of September 2026, the answer is still no for photos, paintings and illustrations. Claude has no built-in image generator, so a request for a painting of a cat in a spacesuit won't come back as a painting.
What Claude does make is visual content built from code. Anthropic calls these custom visuals: flowcharts, interactive charts, side-by-side comparisons and diagrams, rendered right in the chat. They're built with HTML and SVG, the code web pages are written in, so you can click buttons, move sliders and ask Claude to change them. That makes Claude for images a good fit when the picture is really information, and a poor fit when it's a scene.
When Claude hands back an actual photo or illustration, a connected service made it. In that setup Claude writes the request, and the other service's image model draws the picture.
What Claude can do with pictures
Claude AI for images covers five jobs, and each has its own limits.
Build diagrams, charts and SVG visuals
Custom visuals are in beta for all Claude users, in chat and Cowork, on the web and desktop apps. You don't switch anything on. Ask in plain words, such as "turn these steps into a flowchart" or "compare these two plans side by side", or attach a spreadsheet saved as CSV and ask for a chart of it.
- Keeping one. A custom visual lives inside Claude's reply and isn't saved on its own. Use Copy as image for a quick snapshot, Download for an .svg or .html file, or Save as artifact to keep and share it.
- Phones. Custom visuals don't render in the Claude apps for iOS or Android.
- Model choice. Anthropic says Opus does best at visualization tasks, so pick it for anything with many parts.
A downloaded SVG is a vector file, so a flowchart stays sharp at any size and opens in most design programs.
Read and describe your images
Claude can look at a photo, screenshot or scanned page, describe it and answer questions about it. It accepts JPEG, PNG, GIF and WebP. In a chat, click the "+" button at the lower left, choose "Add files or photos", and pick up to 20 files, or paste an image from your clipboard. Images can be up to 8000 x 8000 pixels, and Anthropic recommends at least 1000 x 1000 for reliable reading.
Write prompts for an image generator
Claude is useful before the picture exists. Tell it what the image is for, who will see it and the shape you need, and ask for a finished image prompt. For example: "Write an image prompt for a wide banner for a spring plant sale at a garden center. Include the setting, the light, the colors and a clear area for text." A good result reads like this, and you can paste it into any image generator:
A wide 16:9 photograph of wooden garden center tables crowded with potted herbs, tulips and small lemon trees, shot at eye level in soft morning sun. Warm greens and terracotta, gentle depth of field, and an empty strip of pale sky at the top for a headline. No text, no logos, no people.Make files and PNG charts from your data
With code execution and file creation, Claude produces Word documents, PowerPoint decks, Excel spreadsheets and PDFs, and it can save a data chart as a PNG image. It's on by default for Free, Pro and Max plans. If it's off, open Settings, then Capabilities, and switch on "Code execution and file creation". This is the closest Claude gets to making an image file, and Anthropic describes it for data visualizations, not pictures.
Lay out designs and prototypes in Claude Design
Claude Design, a beta on Pro, Max, Team and Enterprise plans, builds designs, clickable prototypes and one-pagers from a chat. You can upload screenshots, wireframes or pictures you like as a starting point. The finished design exports as PDF, PPTX, standalone HTML or a .zip file, or goes straight to Canva. It opens at claude.ai/design, or from any chat under Output, then Design. On the iOS and Android apps you can view a design, but editing needs the web or desktop app. Images you upload are reference material, and the result is a layout, not a new photo.
Can Claude make images like ChatGPT and Gemini?
No. Claude for generating images isn't an option, while ChatGPT and Gemini both draw pictures from a prompt and edit uploads. As of September 2026, the three compare like this:
| Question | Claude | ChatGPT | Gemini |
|---|---|---|---|
| Draws photos and illustrations from a prompt | No | Yes, with ChatGPT Images | Yes, with Nano Banana 2 |
| Edits your photo from an instruction | Only through a connector such as Adobe | Yes, select an area and describe the change | Yes, for users 18 and older |
| Diagrams and charts | Yes, custom visuals in HTML and SVG, or PNG charts | Yes, as generated images | Yes, as generated images |
| Free plan | Yes, without image generation | Yes, with limited image creation | Yes, with usage limits |
| Starting paid price (US) | Pro, $17 per month billed yearly or $20 monthly | Go, $8 per month | Google AI Plus, $4.99 per month |
For the step-by-step in the other two, read how to make images with ChatGPT and how to use Gemini's image generator. Copilot and Meta AI also draw pictures; the Copilot and Meta AI image guide has their steps.
Can Claude edit photos?
Claude can't change the pixels of a photo itself. It can direct Adobe's tools to do it once you add the Adobe connector, which brings in Photoshop, Lightroom, Firefly, Adobe Express and other Adobe apps. The connector's listing names these jobs:
- Color and light. Adjust color and lighting on one photo or a batch.
- Backgrounds. Blur the background or remove it.
- Framing. Straighten, crop or expand a photo, and resize it for social platforms.
- Designs. Find templates, then swap text and images and change colors in a live editor.
The Canva connector works the same way for design: Claude can search your Canva designs, create new ones, autofill templates and export them. To add either one:
- In a chat, click the "+" button at the lower left, hover over Connectors, and select "Manage connectors".
- Next to Connectors, click the "+" button, then pick Adobe or Canva from the directory.
- Click Connect and follow the sign-in prompts to give Claude access to your account.
Connectors from the directory work on every Claude plan, on the web, desktop and mobile. Connected services handle your files on their own systems and under their own terms, so review each result in the Adobe or Canva app before you use it.
Image connectors: how Claude gets pictures from other tools
A connector can also hand Claude an outside image model. Hugging Face's own guide (August 2025, updated October 2025), for example, describes a connector that lets Claude run image models hosted there, such as FLUX.1 Krea and Qwen-Image. You add it in Claude as a custom connector and sign in with a Hugging Face account, which comes with a limited amount of free usage. Claude writes the prompt and can look at the result, but the picture is made by that model, under Hugging Face's limits and terms.
Two rules from Claude's help center apply. Custom connectors link Claude to services Anthropic hasn't verified, so connect only ones you trust. On the Free plan you can add just one custom connector. Hugging Face has already changed its setup steps once since publishing them, so follow its current instructions.
Where Claude falls short with images
- No scenes or portraits. A product shot, a birthday invitation illustration or a portrait has to come from an image generator.
- Custom visuals are a beta. Quality and complexity vary, and Claude doesn't always choose to draw one. Ask for one directly, for example "make this a diagram".
- Visuals stay off phones. Custom visuals only appear on the web and desktop apps, and shared chats show them only to signed-in viewers on those apps.
- Reading has limits. Small, low-resolution images read poorly. In Word and other non-PDF documents Claude reads only the text, so upload a picture on its own.
- Connectors add accounts. Adobe, Canva and Hugging Face work happens in those services, under their plans and terms.
For a wider tour of what different AI tools make, from songs to videos, see what AI can make for you.
Turn Claude's prompt into a picture in Image Generator
A prompt Claude writes for you can go straight into Image Generator in Toybox AI. Ask Claude to write it for one of Image Generator's five shapes (1:1, 3:4, 4:3, 9:16 or 16:9) so the layout fits, then paste it into "Describe your image" and pick the same shape under "Size". You can attach three reference photos at most if you have a look in mind, and if the image includes words, turn on "Enhanced quality" in "More options" (currently +45 credits). Pictures currently cost 50 credits each, so the 20 free credits given at sign-up won't cover one; look over plans and top-ups before you begin.
When the photo exists and only needs changes, Smart Editor edits that one upload from what you type into "What should change?". It needs Pro or higher, and an HD edit currently takes 85 credits. Smart Editor re-renders the whole image on every edit, so ask it to leave everything else alone, then compare faces and text with your original.