Images in Claude: What It Reads, and Why It Won't Paint

There is a clean split worth learning up front. Claude is genuinely good at looking at images you give it — reading, describing, and reasoning over them. Claude cannot create an image from a text prompt. If you came expecting a "type a sentence, get a picture" tool, you want something else, and this guide points you there.

What Claude does with images: it reads them

Claude has vision. You can drop a photo, screenshot, diagram, or PDF into any current surface — claude.ai, the desktop app, or the mobile apps — and ask about it in plain language. This works across the model family (Opus 4.8, the Claude 5 models, and the faster Haiku 4.5 tier), though exact per-model limits change, so check the current model list at anthropic.com.

Photos and real-world things

Hand Claude a photo of a whiteboard, a plant, a nutrition label, a broken part, or a handwritten note, and it can transcribe, identify, and explain. Ask follow-ups: "what's the total on this receipt," "is this rash-looking thing worth a doctor" (it will hedge, and should), "what tool is this." It reasons about the picture — it does not edit or redraw it.

Screenshots, diagrams, and UI

This is one of the most practical uses. Paste a screenshot of an error message, a spreadsheet, a chart, or an app screen, and Claude can read the text, explain what's happening, and suggest a fix. Feed it a flowchart or architecture diagram and it will walk through the logic. For anything with small text, send the highest-resolution crop you can.

PDFs and documents

Claude reads PDFs — including scanned ones and pages that mix text with figures — and can summarize, extract tables, answer questions, or pull specific clauses. Combined with very large context (up to about 1M tokens on some models, as of July 2026 — confirm current limits), you can load a long report and interrogate it. For files you reuse constantly, put them in a Project so the context persists.

The honest gap: Claude has no image generator

Say it plainly: Claude does not paint pictures. There is no built-in DALL-E, Imagen, or Midjourney equivalent inside it. Ask for "a watercolor fox" and you will not get one. This is a real, current limitation as of July 2026 — not a setting you've missed.

Making visuals anyway — with code and Artifacts

Claude can still produce visuals, just by a different route: it writes code that draws them. Using Artifacts (a live, side-by-side rendered pane), Claude can build charts, SVG diagrams, flowcharts, HTML/CSS mockups, and small interactive tools you can see and tweak on the spot. That covers a surprising amount of "I need a picture" work — as long as the picture is structured (data, shapes, layout) rather than painterly (photos, art, textured scenes).

Rule of thumb: structured and diagrammatic (Claude can build it in code); photographic or artistic (Claude cannot).

What to use when you need generated images

Prices and features on those tools move quickly — check each vendor's current plans before committing.

When Claude is the wrong tool for images

Reach for something else when:

Where Claude wins is the thinking around images: reading them carefully, cross-referencing them with a long document, and turning data into a clean diagram. Where it loses is making the picture itself.

Quick checklist

Related