Pictures and Video with Claude: What It Sees, What It Makes, and the Workarounds
Claude knows how to look: a photo of a crack, a hand sketch, a screenshot of an error or a PDF full of drawings. It does not paint pictures like a camera, but it builds SVG, charts and animations in code, and through connectors such as Canva and Figma it can produce real images too. Video? There is a simple workaround.
01The analogy: a sharp-eyed engineer who drafts but has no camera
Picture a senior engineer who glances at a site photo and immediately tells you what is going on. Ask him for a new picture, though, and he will not pull out a camera. He will sit down at CAD and draft it, or call a photographer. Claude is the same: an excellent eye, drawing through code, and for real photos it calls an outside tool through a connector.
02How it works: from picture to deliverable
- 1UploadDrag or paste a photo, screenshot, sketch or PDF into the chat
- 2Claude looksReads text, numbers, shapes, lines and details in the image
- 3Ask for outputA description, a checklist, a table, code or a clean drawing
- 4Claude buildsText, SVG or HTML in an artifact, or an image via a connector like Canva
- 5DownloadSave as SVG, PNG, PDF or HTML and keep working
03What Claude can read from an image
Formats and limits
JPEG, PNG, GIF and WebP, up to 8000 by 8000 pixels and up to 20 files per chat. Claude understands images but does not generate them itself.
Site photos
Cracks, damp, rebar diameters, anchors and drill holes. It describes and suggests checks; it does not diagnose.
Sketches and drawings
A hand sketch, a scanned plan or a detail inside a PDF. It reads dimensions, labels and symbols.
Screenshots
An error message in Excel, in calculation software or in code, or a whole web page layout.
Tables and charts
A table from a manufacturer catalog or a chart from a report. Ask to turn it into an Excel table.
PDFs with images
Per the help center, for PDFs of up to 100 pages Claude analyzes images and charts too. Beyond that, text only.
04What Claude can create, and how
SVG drawings and diagrams
Claude writes SVG, a vector format that opens in a browser and imports into Illustrator or Inkscape. Ideal for details, sections and flowcharts.
Charts and graphs
A chart from numbers you paste, as an artifact or a visual in the chat, and you can ask for changes.
Animations
An HTML, CSS and JavaScript animation, for example the steps of an anchor installation. Runs in a browser, no video file needed.
Real images via Canva
The Canva connector can create designs and AI images and export them. Requires a Canva account.
Images via Figma or Hugging Face
Figma offers an image generation tool that uses Figma AI credits. On Hugging Face you can run image models in Spaces.
Adobe
The Adobe for creativity connector, announced in April 2026, links Claude to tools across Photoshop, Premiere and Express.
05Example: a photo of a cracked concrete wall
06Example: a hand sketch becomes a clean drawing
07Example: a site video becomes a photo report
08Example: an animated explainer of a chemical anchor installation
09What I want, and how to do it
| I want to | How / which tool | Notes |
|---|---|---|
| Analyze a site photo | Upload it to the chat and ask a focused question | Several angles and a ruler in frame help a lot |
| Fix an error from a screenshot | Paste the screenshot and ask for an explanation and fix | Add the error text too if you can |
| Turn a sketch into a clean drawing | Artifact with SVG | Check every dimension against the original |
| A chart or diagram | Artifact or an in-chat visual | Paste the actual numbers, not just a picture of a chart |
| A real photo or illustration | Canva, Figma, Adobe or Hugging Face connector | Claude alone does not generate photos; may use credits |
| Understand a video | Frames from the video plus the transcript | In Cowork, frames can be extracted with ffmpeg |
| Cut or convert a video | Cowork or Claude Code with ffmpeg | Work on a copy, not the original |
| An animated explainer | HTML animation in an artifact, or Canva and Figma | Figma can export animation as MP4 |
| My image inside a site or artifact | Upload the file and ask to embed it | In a plain chat you may need a link to a hosted image |
10Get started in five minutes
- 1
Take a good photo
Sharp, well lit, not at a steep angle. For small details, shoot one close-up and one wider shot for context.
- 2
Drag it into the chat
Drag the file, paste a screenshot with Ctrl+V, or click the attach button. You can add several images at once.
- 3
Ask a focused question
Not just what is this, but for example: what is the approximate crack width and what should I check. Add what you know: material, building age, location.
- 4
Ask for a deliverable
A table, a checklist, SVG, a chart or code. If you need a real image, connect Canva or Figma in the connectors settings.
- 5
Check and save
Compare numbers and dimensions with the source, then download the file from the artifact. Links from outside tools can expire, so keep a copy.
11Claude vs ChatGPT and Gemini for images and video
Claude
- Reads images, screenshots and PDFs with drawings
- Does not generate photos or illustrations itself
- Excels at SVG, charts, diagrams and animations in code
- AI images through connectors like Canva, Figma and Adobe
- Video: via frames and transcripts, or ffmpeg in Cowork
ChatGPT and Gemini
- They read images too
- ChatGPT creates and edits images directly, including on the free plan
- Gemini creates images, and short videos on paid plans
- Handy when you need a quick rendering or illustration
- For precise drawings with dimensions, code and SVG still win
12FAQ
Can Claude draw me a picture?
Not like an image generator. Anthropic states plainly that it is an image understanding model only. It does write SVG, HTML and charts, and through a Canva, Figma or Hugging Face connector you can get a real image.
Can I upload a video?
At the moment Claude does not watch video files. Upload a few frames and the transcript instead. In Cowork or Claude Code it can extract the frames itself with ffmpeg.
How far can I trust dimensions it reads from a photo?
Carefully. The docs note that positions and counts are approximate, and that small, blurry or rotated images cause mistakes. Any number that goes into a calculation or report gets checked against the source.
What about privacy?
Before uploading, blur faces, license plates, client names and addresses. Claude also does not identify people by their faces. For client work, check your organization's policy.
And copyright?
Do not upload other people's images in order to copy them, and check the terms of the tool that generated an image before commercial use. A drawing rebuilt from your own sketch is yours.
How do I put my own picture into a site Claude builds?
Upload the picture and ask Claude to embed it. In Cowork and Claude Code the file sits in your folder and it places it in the page. In a plain chat, the safe route is to give a link to an image already hosted on your site.