The problem
Why church records are so hard to read
Church and parish registers are among the most valuable genealogical sources — and among the most difficult to use. A single register can span decades of baptisms, marriages, and burials, written by multiple pastors in scripts that changed over centuries. German records switch from Latin to Kurrent to Sütterlin. French registers use Ronde and Coulee before the Revolution. Swedish kyrkoböcker mix formal and informal hands across hundreds of pages.
Multiple scripts in one register — Latin entries, German Kurrent marginalia, French annotations
Abbreviations everywhere: 'b.' for baptised, 'd.d.' for died, Latin shorthand for sacraments
Faded ink, bleed-through, and water damage on paper that's 200-400 years old
Unfamiliar letterforms: Kurrent 'e' looks like 'n', Sütterlin 's' looks like 'f'
No standardised spelling — the same name appears five different ways
The solution
AI that reads parish registers across centuries and languages
Transkribus uses handwritten text recognition (HTR) — AI trained on millions of historical handwriting samples. Unlike standard OCR, which fails on cursive and connected scripts, Transkribus learns how each script type works: the ligatures of Kurrent, the abbreviations of Latin, the flourishes of 17th-century French. Upload a photo or scan of your church record, select a model, and get searchable text back in minutes.
300+ public models — including models trained specifically on church records and parish registers
Reads German Kurrent, Sütterlin, Fraktur, Latin, French, Swedish, Dutch, Polish, and 100+ more
Handles damaged, faded, and low-contrast documents that defeat standard OCR
Confidence scores on every line — so you know which words to double-check
Export as plain text, PDF, or structured XML for your genealogy database
Addres to dear Isabella on the Authors
recovery
O Isa pain did visit me
I was at the last extremity
How often did I think of you
I wished your graceful form to view
To clasp you in my weak embrace
Indeed I thought Id run my race
Good Care Im sure was of me taken
But indeed I was much shaken
At last I daily strength did gain
Comparison
Transkribus HTR vs. Standard OCR on Church Records
Church and parish registers present unique challenges that standard OCR was never designed to handle. Here is how Transkribus HTR compares when processing historical church records.
| Feature | Transkribus HTR | Standard OCR |
|---|---|---|
| Historical cursive scripts | Reads Kurrent, Sütterlin, Secretary Hand, Copperplate, and 100+ historical scripts | Designed for printed text only — fails on cursive and connected handwriting |
| Latin abbreviations | Models trained on real parish registers understand common Latin shorthand (b. for baptised, d.d. for died) | Cannot interpret manuscript abbreviations or ligatures |
| Mixed scripts in one register | Handles pages mixing Latin headings, Kurrent body text, and French marginalia | Requires uniform typeface — mixed scripts cause garbled output |
| Damaged and faded documents | Trained on real historical documents with ink bleed, water damage, and foxing | Performance degrades significantly on low-contrast or damaged pages |
| Tabular parish register layouts | Layout analysis detects columns, rows, and marginal notes in structured registers | Assumes simple left-to-right text flow — misreads columnar layouts |
| Custom model training | Train a model on your specific pastor’s handwriting with 50–100 pages of ground truth | Fixed models — cannot be adapted to specific handwriting styles |
| Confidence scoring | Per-line confidence scores highlight uncertain readings for targeted review | Varies — often absent or unreliable on handwritten text |
Comparison reflects general capabilities of HTR (Transkribus) versus standard OCR engines on handwritten church records. Results depend on document condition, script type, and model selection.
How to transcribe a church record in 4 steps
Upload your document
Take a photo with your phone or upload a scan. Transkribus accepts JPG, PNG, PDF, and TIFF files.
Select an AI model
Choose from 300+ public models. Filter by language, century, and script type to find the best match for your record.
Run text recognition
Click 'Recognise' and let the AI work. A single page takes about 30 seconds. Entire registers can be processed in batch.
Review and export
Check the transcription against the original image. Fix any errors in the built-in editor, then export as text, PDF, or XML.
Models for church records
AI models trained on parish registers and church books
Transkribus's public model catalog includes models built specifically for the scripts and document types found in church records. Many were trained by genealogists and archivists on real parish register collections, so they understand the abbreviations, column layouts, and script variations that generic OCR tools miss entirely.
German church records: Kurrent and Sütterlin models covering 1600s–1940s
Latin records: Models trained on Catholic parish registers across Europe
French records: Ancien Régime and post-Revolution parish registers
Swedish kyrkoböcker: Models for 18th and 19th-century church books
Dutch, Polish, Czech, and more — new models added by the community regularly
Custom training
Train a model on your specific church records
If your records use an unusual hand or dialect that public models don't handle well, you can train a custom model on your own data. Transcribe 50–100 pages as ground truth, and Transkribus trains a model optimised for exactly your document type. The model improves as you add more training data — and you can share it with the community if you choose.
Start with as few as 50 transcribed pages
Fine-tune on your specific pastor's handwriting or regional script variant
Combine with public models for best results on mixed collections
Share your model with other genealogists working on similar records
Frequently Asked Questions
Accuracy depends on the script, condition, and model used. On well-preserved German Kurrent church records, our best models achieve 95%+ character accuracy (under 5% Character Error Rate). Damaged or unusual scripts may need a custom-trained model to reach the same level. Every line comes with a confidence score so you can see which words need checking.
Yes. Many European church records — especially Catholic baptism, marriage, and burial registers — are written in Latin. Transkribus has public models trained on Latin handwriting from the 16th to 19th centuries. The AI handles common Latin abbreviations and ligatures found in parish registers.
Transkribus gives you 50 free credits every month — enough to transcribe about 50 pages. No credit card required. If you need more, paid plans start at affordable rates. See our plans and pricing page for details.
Yes. Transkribus accepts photos from smartphones as well as high-resolution scans. For best results, take the photo in good lighting with the document flat and fill as much of the frame as possible. JPG, PNG, PDF, and TIFF formats are all supported.
You can train a custom model on your own data. Transcribe 50–100 pages of your records using the built-in editor, then let Transkribus train a model specifically on your documents. This works well for unusual scripts, regional variants, or particularly difficult handwriting.
Yes. You can upload a full register as a multi-page PDF or as individual images, then batch-process all pages with a single click. Transkribus handles the layout detection (finding text lines, columns, and table structures) automatically.
You can export transcriptions as plain text, searchable PDF, DOCX, PAGE XML, ALTO XML, TEI-XML, or CSV. The structured XML formats preserve coordinates and layout information, which is useful if you're building a database or digital edition.
Related guides
Explore more genealogy record types
Church records are just one piece of the puzzle. Explore our other record-type guides: Census records · Vital records · Military records · Wills & probate · Immigration records · Land deeds
Ready to unlock your church records?
Create a free account and start transcribing parish registers today. 50 free credits every month — no credit card required.
50 free credits every month — No credit card required
200M+Pages processed
500K+Users worldwide
300+Public AI models