Scans to text, layout kept.
Send a scanned PDF or a photo of a page. OCR returns the text in reading order, tables split into rows and cells and checkboxes marked as ticked or empty. It runs on our own server in Slovenia; documents are not stored after the answer.
from €45 / month for 1000 pages · or per page from credit
PODATKI O GRADNJI
naziv gradnje ČISTILNA NAPRAVA MEJA
VRSTE GRADNJE ☑ NOVOGRADNJA - NOVOZGRAJEN OBJEKT
☐ NOVOGRADNJA - PRIZIDAVA
☐ REKONSTRUKCIJA
Text a program can work with
Not just a string of characters: every page comes split into blocks in reading order, each with its position on the page, so tables, titles and footers do not run together.
Blocks in reading order
Title (with its level), header, footer, text, table, image, image caption, reference, page number. Each block has its position on the page.
Tables by cell
A table comes as rows and cells and as HTML, ready for a spreadsheet or a database.
Checkboxes
Ticked and empty boxes are marked (☑ and ☐), so a form tells you what was chosen.
Drawings apart
A plan or a detail is recognised as a drawing; its labels stay out of the document text.
One call per document
The request body is a PDF, one page image or JSON with a base64 file. Up to 60 MB and 50 pages per request; ?pages=2-7 picks pages, ?format=text returns plain text only.
# pages 1 to 5, answer in JSON $ curl -X POST "https://ocr.mediaatlas.si/v1/ocr?pages=1-5" \ -H "Authorization: Bearer $KEY" \ -H "Content-Type: application/pdf" --data-binary @document.pdf {"model": "unlimited-ocr-2", "page_count": 6, "pages": [{"page": 1, "blocks": […], "drawing": false}], "text": "…"}
Read, then forgotten
On our server in Slovenia. The model runs with us, not with a foreign provider. The document goes nowhere else.
Gone after the answer. We do not store documents and do not train models on them.
An open model. Unlimited-OCR-2 (MIT licence) reads one page at a time at 300 dpi.
What people ask before the first document
How much does it cost?
A plan is €45 a month for 1000 pages. Or pay per page from credit, with no plan. Prices exclude VAT. The key and the purchase are on your MediaAtlas account at services.mediaatlas.si; the same key works for every one of our services.
Do you keep the documents?
No. The document is read on our own server in Slovenia and not stored after the answer. We do not train models on your documents.
What can I send, and how much at once?
A PDF or one page image (PNG, JPEG, WebP, HEIC from iPhones, TIFF, BMP), up to 60 MB and 50 pages per request. Send longer documents in page ranges.
What about phone photos?
A photo is turned upright by its orientation tag and straightened when the text is tilted, so a receipt photographed at an angle still keeps its prices and total. Each page says by how many degrees it was turned.
What happens to plans and drawings?
A page where images cover 30 % or more of it is marked as a drawing (a plan or a detail, for example). Its labels are left out of the document text so they do not mix with the content.