Separate Photos from a Scrapbook Page

One page in, separate photos out

You send

You get back

Enlarge

What this does

Scan the page the way it sits in the book, with its borders, captions and corner mounts still on it. We read the page, find each photograph on it, and cut each one into a separate image. Every one of those images then goes through the same restoration the rest of the site runs, on its own.

What comes back is the photographs, as separate files, not the page they were mounted on. If you would rather restore the page as one picture, order it as photo restoration instead.

What to send us

• PDFs are fine. Flatbed scanners default to PDF, so a scanned album very often arrives as one. Each page of the PDF becomes its own page scan, which is exactly the right shape for this.
• Images work too. JPEG, PNG, HEIC and WebP. A phone photo of a page can work, but a scan is better, because glare and a tilted angle both cost you at the cropping step.
• Up to 20 pages per order. Reading a page takes real time, so the limit sits well below the normal order size. A larger album goes out across more than one order.
• One page per scan, the whole page in frame. The prep guide covers lighting and framing. Everything in it applies here, except the part about cropping the borders out. Leave those on.

How it works

1. 1

   Scan the page

   One scan per album page, with the whole page in frame. A PDF straight from your scanner is fine. Each page of it becomes its own page scan, in your browser, before anything is uploaded.

2. 2

   We look for photographs

   We check your pages for photographs, then show a photo-count summary and your order total before payment. There is no image or crop preview, and detection can miss photos or get their edges wrong.

3. 3

   Each photograph is cropped out

   After payment, each detected photograph is cropped into its own image. We leave a small margin around the detected edges to reduce the chance of cutting into the picture. The restoration step then aims to straighten the photo and trim away the album page, corner mounts, and border.

4. 4

   Each one is restored on its own

   Your chosen restoration level runs on every cropped photograph separately, so a faded snapshot beside a sharp one each get what they need. You get the separate photographs back, not the page.

What it costs

No other fees: the page price is the whole price.

A page is billed flat, whatever is on it. One page price covers every photograph we find on that page, up to 10. A page with two photographs costs the same as a page with eight. You see the total after selecting your files and restoration level, before uploading.

Loose photographs, with no page to separate, are priced per photo on the pricing page.

What it does not do

Worth reading before you scan, because some of these decide whether your pages are a good fit at all.

• ✗Separation is designed to find photographs, not caption cards or documents. Those items may be left out of the returned crops. Keep your original scans to preserve the page layout and captions.
• ✗The restoration step aims to crop away corner mounts, album page, and the print's own border, and straighten tilted prints. The AI judges where each picture ends, so unusual frames such as Polaroids, studio mount cards, or deckled edges may be trimmed more or less than expected.
• ✗You cannot preview or adjust the crops before payment. You can review the count summary and price, or start over with a different selection or without separation.
• ✗If no separate photographs are found, the page is restored whole at the same price. We separate at most 10 detected photos per page; additional photos are not included.
• ✗Each order is limited to 20 pages and 100 images to restore in total. If detection puts your order over that image limit, split it into smaller orders before paying.

What we have measured

We have run this over four real album pages. It found 15 photographs on them: six on the first page, one on the second, six on the third, two on the fourth. Nothing on the cream page backgrounds was mistaken for a photograph, no caption was picked up as one, and a studio portrait that filled most of a page came back as a single photograph rather than being missed.

That is the evidence we have. It is four pages, not a benchmark, so we are not going to put a percentage on it. It works well on pages like those, and when a page defeats it, the page is restored whole rather than the order failing.