AI Book Scanning: Point, Shoot, Catalog

Take three photographs. Get thirty-one fields. Sit down because that's genuinely all it takes.

What It Does

AI Book Scanning uses computer vision to extract bibliographic metadata from photographs of your book. You photograph the title page, optionally the cover and colophon, and Shelvd sends the images to an AI model that reads them — not with OCR, but with genuine understanding of what it's looking at.

The result is a draft book record with up to 31 fields pre-filled: title, subtitle, author, publisher, publication year, edition, binding type, format, language, ISBN, condition observations, and more. You review the draft, adjust anything the AI got wrong (it will occasionally hallucinate a publisher or invent a subtitle — it's AI, not a librarian), and save.

The entire process takes about fifteen seconds. Which is less time than it takes to type "Tolkien, J.R.R." in the contributor field.

The Three Photos

You can submit one to three photographs per scan. The AI gets smarter with more context.

Photo 1: Title Page (required) The bibliographic source of truth. This is where the AI finds the title, author, publisher, place of publication, and year. If you photograph nothing else, photograph this.

Photo 2: Cover (recommended) The front cover tells the AI about binding material, cover type, dust jacket presence, and sometimes the series. It's also where signed copies reveal themselves.

Photo 3: Colophon (recommended) The colophon — usually at the back of the book, sometimes at the front in fine press editions — contains edition statements, limitation numbers, printer details, and paper information. For antiquarian books, this is where the real metadata lives.

Tips for Better Photos

Good lighting. Natural daylight or a desk lamp. No flash — it creates glare on glossy pages.
Flat and square. Lay the book open, hold the page flat. The AI can handle some skew, but a straight-on photo yields better results.
Full page visible. Don't crop too tight. The AI needs to see the margins to understand the layout.
Readable text. If you can't read it in the photo, neither can the AI. Blurry photos produce blurry results.

What Gets Extracted

The AI attempts to extract all of the following from your photographs:

From the title page: title, subtitle, original title, series, series number, contributors (with roles), publisher, publication place, publication year, language, ISBN.

From the cover: cover type, binding material, dust jacket presence, signature status, condition observations.

From the colophon: edition, impression, issue/state, printer, printing place, paper type, page count, format, limitation notes.

Always attempted: a condition assessment based on what's visible in all provided photos — foxing, toning, tears, bumped corners, and other defects the AI can spot.

Not every field will be populated. A modern paperback's title page might yield fifteen fields; a seventeenth-century folio with a detailed colophon might yield twenty-five. The AI extracts what it can see and leaves the rest blank — which is the honest approach.

The Scan Flow

Go to Scan from the navigation (or /books/scan)
Your remaining AI Scans balance is shown at the top
Upload one to three photos — use the camera or select files
Click Scan
Wait roughly ten seconds while the AI examines your photos
Review the extracted data in a preview panel
Edit any fields that need correction
Click Save to Library to create the book

The saved book arrives as a draft with status "in collection." All photos are stored in the book's image gallery with appropriate labels (cover, title page, colophon), so you don't need to re-upload them.

When the AI Gets It Wrong

It will. Not often, but predictably in certain situations:

Handwritten titles or unusual typefaces may be misread
Multiple title pages (common in bilingual editions) can confuse priority
Reprint publishers sometimes get credited instead of the original publisher
Dates in Roman numerals — the AI handles these surprisingly well, actually, but MDCCLXXVIII can trip up anyone
Condition notes are conservative by design. The AI might describe something as "light foxing" when you'd call it moderate. Better cautious than generous.

Review the draft before saving. Think of the AI as a very fast, very eager cataloging intern who's read a lot of books but has never actually held one.

Credits

Each scan costs one AI Scan credit, regardless of how many photos you include (one photo or three — same cost). Credits come from two sources:

Monthly allowance: Collector Pro gets 10 scans/month, Dealer gets 50 scans/month
Purchase packs: 25 scans for €2.50, 100 for €7.50, 500 for €29

Monthly credits reset on the first of each month. Purchased credits never expire. Shelvd uses your monthly credits first, then draws from purchased credits.

Your current balance is always visible at the top of the Scan page and in Settings → Billing.

For everything about credits, see AI Scan Credits.

Who Can Use It

AI Book Scanning requires the Collector Pro or Dealer tier. Collector (free) users don't see the Scan option in the navigation. This isn't arbitrary gate-keeping — each scan costs real money on Shelvd's end (the AI model charges per image analyzed), so it's bundled with paid tiers.

If you're on the free tier and want to try it: upgrade, scan a few books, and see whether pointing a camera beats typing. For most people, it does.

AI Scanning vs. Library Lookup

They're complementary, not competing:

	AI Scanning	Library Lookup
Input	Photos of your copy	ISBN or search terms
Source	What the AI sees	What libraries have on file
Unique to your copy	Yes — condition, signatures, provenance clues	No — canonical metadata only
Best for	Older books, unsigned, no ISBN	Modern books with ISBN
Speed	~15 seconds	~5 seconds
Cost	1 credit per scan	Free

The ideal workflow: scan the book with AI to get a draft, then enrich from a library catalog to fill any gaps the photos didn't reveal. Two passes, one thorough record.

📖 Related on the blog: Cataloging on a Sunday