What you will learn
- Understand plainly what Claude can and cannot do with images
- Analyse a photo, screenshot, or chart you upload
- Write alt text that actually describes an image
- Turn a diagram idea into SVG or diagram code
- Review screenshots for layout and readability problems
- Write briefs for image tools from a clear description
The honest starting point: Claude does not generate images
Plainly stated: Claude does not generate images. If you ask it for a photograph, an illustration, or a picture of a product, it cannot produce one. That is a real limit, and knowing it saves you from building a workflow around a feature that does not exist. What Claude does well with media is different: it can look at images you upload, describe them, analyse them, and write code that draws graphics, such as SVG illustrations and diagrams, which you can then view in an artifact or a browser.
This page teaches that real set of skills. You will learn to get useful analysis out of images you already have, to write descriptions and briefs that other tools can use, and to create diagrams and graphics as code. Each of those is genuinely useful, and none of them pretends Claude is an image generator.
Analyse an image you upload
Upload an image and ask a specific question about it, rather than a general "what is this?" Ask what the chart shows, whether the screenshot contains an error message, what objects appear in a photo, or how a design could be clearer. Specific questions get specific answers; general questions get general descriptions.
For charts and graphs, ask Claude to read the values, name the trend, and flag anything misleading, such as an axis that does not start at zero or a label that is missing. For screenshots of an app or a website, ask it to list what it sees element by element, so you can check its reading against your own.
Write alt text and descriptions
Alt text exists so people who cannot see an image still understand it, and Claude can draft it well when you tell it the context. Upload the image, say where it will appear and what matters about it, and ask for a description of one or two sentences that covers the essential content. A product photo needs different alt text from a chart: the photo needs what the product looks like, the chart needs what the data says.
Review every description before publishing it. Check that names, numbers, and the main point are right, and cut anything the image does not actually show. Alt text that invents details is worse than short alt text, because readers who rely on it cannot check the image themselves.
Write alt text for this image, in one or two sentences. The image will appear on a product page for a handmade ceramic mug. Describe the mug's colour, shape, and size impression, and mention the handle. Do not invent details that are not visible in the image.
Diagrams and graphics as code
Claude can write SVG code for simple illustrations, icons, and diagrams, and it can write diagram code for flowcharts and structure charts. Describe the diagram you want: the boxes, the labels, the connections between them, and the order they should appear in. The result is code you can open in a browser or paste into an artifact and see rendered.
Start simple. A clean flowchart with five boxes beats a complex diagram with twenty, because each element is easy to check and easy to fix. Ask for one change at a time when refining: move a box, rename a label, or change a colour, and view the result before the next change.
Review screenshots like a checklist
Screenshots are useful for design review. Upload a screenshot of a page or a slide and ask Claude to check readability: is the text large enough, is the contrast clear, is the most important element the most prominent, and is anything overlapping or cut off? Treat the answer as a checklist to verify yourself, not as a final verdict, because fine visual details are easy to misread from an image.
For a slide deck or a set of social posts, review them as a set. Ask whether the style is consistent across the images, whether the message is clear in each one, and whether any image contradicts the others. Consistency problems are easier to spot across a set than one image at a time.
Tip: When you need a generated picture, write the brief with Claude and generate the picture elsewhere. A clear brief, subject, setting, style, and what to avoid, produces far better results in an image tool than a vague one-line request.
Briefs for image generation tools
Since Claude does not generate images, the practical workflow is to use it to write the brief, then run that brief in a dedicated image tool. Give Claude the purpose of the image, the subject, the setting, the style, and anything to avoid, and ask for a tight brief of a few sentences. Also ask for two or three variations, so you have options if the first generation misses.
Keep the brief honest about what the image is for. A product illustration, a blog header, and a social post need different compositions, and saying which one you are making helps Claude write a brief that fits. Check current plans and features in the tools you use, because image capabilities change often.
Frequently asked questions
Can Claude generate images?
No. Claude does not generate images such as photographs or illustrations. It can analyse images you upload, describe them, and write code for SVG graphics and diagrams.
What can Claude do with an image I upload?
It can describe it, read charts, find text or error messages in screenshots, and review layout and readability. Ask specific questions for the most useful answers.
Can Claude write alt text for my images?
Yes. Tell it where the image will appear and what matters about it, and ask for a short description. Review it before publishing, since readers who rely on alt text cannot check the image.
Can Claude make diagrams?
Yes, as code. It can write SVG for simple illustrations and diagram code for flowcharts and structure charts, which you can view in an artifact or a browser.
How do I get an AI-generated picture then?
Write the brief with Claude, subject, setting, style, and what to avoid, and run that brief in a dedicated image generation tool. A clear brief produces much better results.