AI OCR to Markdown
Gukoresha
Urugero
Ikimenyetso cy'amazi
Kwiyandikisha
Urugero:
+ GPT-5, Claude, Gemini
Drop an image, screenshot, or scan and get back clean Markdown — headings stay headings, tables stay tables, lists stay lists. Powered by our self-hosted Unlimited-OCR model, tuned for layout-aware document parsing rather than a flat text dump. Best for structured pages: papers, invoices, forms, reports.
Markdown output
What it's good for
Tables & forms
Invoices, spec sheets, and reports come back as real Markdown tables instead of a jumbled wall of numbers — paste straight into a doc or a RAG pipeline.
Papers & articles
Headings, sections, and lists keep their structure, so a scanned report becomes editable Markdown you can drop into a docs site or feed to an LLM.
Dense & multilingual
The model is built for dense, complex pages and handles Chinese, English, and more — useful where a plain OCR engine garbles tight layouts.
Amahitamo y'Ikirenga
Igisubizo
Hasi
Birenzeho
❤️ Free.ai? Abayobozi!
< a href=" / /" Imisusire=" Ibara: #"> Hejuru Kuri Kubona A Ihuza na Kubona 25,000.
Birenzeho?
Hejuru Kigenga ya: 30K / Umunsi + 10K
Kwiyandikisha
Kubaza...
Convert any image, screenshot, or scan into clean Markdown with free AI OCR. Preserves headings, tables, and lists — powered by the self-hosted Unlimited-OCR model.
Kuri Gukoresha AI OCR to Markdown
1
Iyinjiza
Umwandiko, Kohereza A Idosiye, Cyangwa. Aderesi: Bya ngombwa.
2
Kurema
Kubaza... in amasogonda ikoresha i Gufungura - Inkomoko Udushushondanga.
3
Iyimura & Gusangiza
, Gukoporora, Cyangwa Gusangiza Igisubizo. ya: Bwite na Ikoresha:.
iyi Biturutse
iyi Kuva: Inyandikoporogaramu. - Impera Akadomo, -, Oya Birenzeho Bya ngombwa. BIHUYE i Urubuga Imigaragarire.
curl -X POST https://api.free.ai/v1/pdf/extract/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"file": "@document.pdf"}'Amadosiye
AI OCR to Markdown — FAQ
Upload an image, screenshot, or scan and get back clean Markdown — headings, tables, and lists keep their structure instead of collapsing into a flat text dump. It is powered by Unlimited-OCR, a self-hosted 3B model tuned for layout-aware document parsing, so the output is ready to paste into a docs site, a note app, or an LLM pipeline.
Plain OCR (our /ocr/ tool, Tesseract) gives you the raw words in reading order — great for a quick copy-paste, but tables flatten and headings lose their hierarchy. OCR to Markdown understands the page layout and reconstructs it: a table becomes a Markdown pipe-table, a heading stays a heading, a bullet list stays a list. Use plain OCR for speed, this for structure.
Unlimited-OCR (baidu/Unlimited-OCR), an open MIT-licensed 3B model built on DeepSeek-OCR. We run it on our own GPUs — nothing is sent to a third-party API. It is purpose-built for one-shot document parsing rather than single-line text recognition.
Yes — tables come back as proper Markdown pipe-tables, which is the whole point of using this over plain OCR. Very complex nested or multi-column tables flatten more aggressively (a fundamental Markdown limitation), but standard invoices, spec sheets, and report tables convert cleanly.
It is multilingual with particularly strong Chinese and English support, and handles many other scripts. Accuracy is best on clear, high-resolution pages; very low-resolution or heavily stylized text is harder, as with any OCR.
That tool takes a whole PDF and runs IBM Granite-Docling page by page — best when you already have a multi-page PDF. AI OCR to Markdown takes a single image or screenshot and runs the larger Unlimited-OCR model, which is stronger on dense or complex single pages. Use the PDF tool for documents you have as PDFs; use this for screenshots, photos of pages, and individual scans.
AI OCR to Markdown takes one image (PNG / JPG / WebP) at a time — one page per image. For multi-page PDFs, use /document/pdf-to-markdown/ (whole-PDF Markdown) or /ocr/document/ (searchable-PDF / text). If you only have a PDF page, export or screenshot the page and drop it here.
If you do not want Markdown structure — just the raw text in reading order — pick Plain text. The model then runs in straight-OCR mode and returns the words without headings, tables, or list formatting. Markdown is the default because it is what makes AI OCR to Markdown distinct.
It works on phone photos of paper, but clean, flat, well-lit shots convert best. Glare, heavy skew, and shadows hurt accuracy. If a photo comes out poorly, re-shoot it straight-on with even lighting, or scan at 200+ DPI.
Images are processed on our GPUs and discarded — we do not train on your uploads. We log only a timestamp and token cost for billing, never the document content. Signed-in users get a generation-history entry (a short preview) which you can delete anytime from /account/?tab=history. Privacy policy at /privacy/.
Free in the daily token pool (~800 tokens per page) — no credit card. Your first conversions work without an account; signing up unlocks the full daily pool, history, and access across all of Free.ai's 400+ tools.
Yes. POST multipart to /v1/ocr/ with your image, model=unlimited-ocr, and output_format=markdown (or plain). The response JSON has a text field with the Markdown. Bearer auth, 10K tokens/month free. /api/ has the curl example.
iyi