How to Chat With Images and Scanned Documents Using AI: A Step-by-Step Guide

To chat with images and scanned documents, a tool first looks at the picture — a photo of a whiteboard, a scanned contract saved as a JPG, a screenshot of a chart — and produces a caption or transcription of what's in it. From there you can ask about it in plain language: “What does this scanned page say about the deadline?” or “Summarize the diagram in this photo,” instead of retyping the text yourself or zooming in on a blurry image looking for one detail.

Under the hood it works the same way as chatting with any other document: once the image has been described in text, that text is split into chunks, each chunk is embedded, and your question retrieves the most relevant ones before a language model writes an answer from them. We cover the general mechanism in our RAG explainer; for an image the only difference is the extra captioning step that turns a picture into searchable text in the first place.

1. Pick a tool that accepts standalone images (JPG, PNG) as a source, not just PDFs and text files.
2. Upload the photo or scan and wait a few seconds while it's captioned and indexed.
3. Ask a broad question first, like “What is shown in this image?”, to confirm it read the picture correctly.
4. Follow up with specific questions — a figure from a scanned table, a label on a diagram, a line of text from a photographed page.
5. Add related files to the same project so a single question can draw on the image and your other sources together.

An image-captioning model reads a picture the way a person skimming it would: it transcribes printed or typed text reasonably well, describes charts and diagrams, and picks up labels and headings. It's less reliable on messy handwriting, and a dark, blurry, or heavily angled photo gives it less to work with than a flat, well-lit one. The practical takeaway is simple — a clear photo taken straight-on, or a proper scan instead of a hasty phone snap, turns into a far more accurate answer than a rushed shot of a page.

This is exactly where the workflow pays off for students: studying with AI from your own notes already means uploading photos of handwritten pages alongside lecture slides, and having both readable in the same project means a question can pull from a photographed page and a typed PDF in one answer. The same captioning step also covers images embedded inside a slide deck, as described in how to chat with Office files — a chart on a slide becomes searchable text too, not just the bullet points around it. Once everything is indexed, generating a quiz from the mixed set tests recall across photos and text together.

Doxylets you add a photo or scanned document the same way you'd chat with any other document: upload it as a source, wait for it to be captioned and indexed, and start asking. It's part of the same AI document chat and document Q&A workflow, so a photographed page sits alongside your PDFs, slides, and web links in one project, with citations back to the source you actually uploaded.