Transkribus Platform

06/08/2026

Great ideas in digital humanities don’t happen in isolation.

They grow when researchers, archivists, developers, and heritage professionals come together to share experiences, discuss challenges, and learn from each other.

That’s what makes the Transkribus User Conference 2026 a special opportunity to connect with the people working with historical documents and digital collections every day.

This year’s programme features practical examples from research projects, archives, and cultural heritage institutions — covering topics such as handwritten text recognition, collection workflows, and new approaches to working with historical sources.

From 21–23 September at the University of Passau, you'll hear how colleagues are tackling real-world challenges, discover practical approaches you can apply to your own work, and exchange ideas with the Transkribus community.

Whether you’re already using Transkribus or simply interested in digital approaches to historical documents, TUC 2026 is a place to connect, learn, and share experiences.

🎟️ Secure your ticket now and join us in Passau:

Join us at the Transkribus User Conference 2026 - Connect, Learn, and Explore the Future of Unlocking Written Heritage with AI

04/08/2026

How do you get 250,000 insect specimen labels into a searchable database?

The Museum für Naturkunde Berlin holds one of the largest natural history collections in Germany, including around 15 million insect specimens.

Each specimen label contains valuable scientific information — collection locations, dates, collectors, and taxonomic details. But when this information exists only on paper, accessing it at scale becomes a major challenge.

How can you make hundreds of thousands of labels searchable without manually entering every single record?

Working with Transkribus, the museum developed a customised workflow for transcribing and structuring its entomological collection.

The result was a structured dataset of 250,000 transcribed and enriched specimen labels, providing researchers with improved access to valuable biodiversity data.

Have a large-scale collection waiting to be digitised? See how the Museum für Naturkunde did it.👇

Transkribus collaborates with the Museum für Naturkunde Berlin to digitise and transcribe 250,000 insect specimen labels, enhancing digital access for researchers and educators.

30/07/2026

AI is everywhere, but do we know what data it was trained on?

The rise of LLMs has sparked important conversations around training data: where it comes from, how it is used, and whether researchers can understand or reproduce the results.

For historical document research, transparency matters. Knowing exactly which data was used to train a model is essential for reproducibility, evaluation, and building on previous work.

With Datasets in Transkribus, researchers can freeze dataset versions during model training, creating a traceable link between training data and custom models. This makes it possible to document workflows and reproduce results.

Researchers remain in control: datasets can stay private or be shared publicly, and custom models can be published with the option to make their underlying training data visible.

Learn more about Datasets: https://eu1.hubs.ly/H0xgqlW0

Interested in the wider discussion around general-purpose vs. specialised AI?

Join us at the Transkribus User Conference 2026 (21–23 September, University Passau), where this topic will be one of the key themes of the conference.

Full programme and registration: https://eu1.hubs.ly/H0xgqFg0

Datasets in Transkribus are curated sets of pages used to train, validate, and test custom AI models.

28/07/2026

Starting a new transcription project? The right setup from day one can make all the difference.

Without a solid understanding of your workflow, tasks like uploading files, organising documents, adjusting page layouts, and running text recognition models can quickly become time-consuming.

Our Beginners Webinar recording is available to help you get started with Transkribus at your own pace. In just one session, you'll learn how to:

- Upload and organise your documents
- Configure page layouts
- Choose and apply text recognition models
- Build an efficient workflow for your projects

Establishing good practices early will save you time, keep your data well organised, and make it easier to scale your research as your collections grow.

Watch the full webinar recording here:

Transkribus is the most popular tools for automatic text recognitio...

22/07/2026

We’re happy to see our Transkribus Teaching Scholarship supporting new learning opportunities and research in the digital humanities!

Z przyjemnością mogę podzielić się informacją, że nasz projekt realizowany na Uniwersytecie Kardynała Stefana Wyszyńskiego w Warszawie otrzymał Transkribus Teacher Scholarship.

W ramach programu otrzymaliśmy dostęp do planu Scholar oraz 1000 kredytów, które zostaną wykorzystane w badaniach oraz podczas zajęć dydaktycznych poświęconych wykorzystaniu sztucznej inteligencji w humanistyce cyfrowej.
W ostatnich miesiącach intensywnie rozwijamy Laboratorium AI dla Humanistyki Cyfrowej. Nasze prace koncentrują się m.in. na:
* rozpoznawaniu pisma ręcznego (HTR),
* analizie historycznych dokumentów,
* wykorzystaniu modeli językowych (LLM),
* budowie baz wiedzy i grafów wiedzy,
* zastosowaniu AI w archiwistyce i badaniach historycznych.
Wsparcie Transkribusa pozwoli nam prowadzić kolejne eksperymenty, rozwijać materiały dydaktyczne oraz angażować studentów w praktyczne projekty związane z cyfrowym opracowaniem źródeł historycznych.

Dziękujemy zespołowi Transkribus, a w szczególności Sarze Mansutti i Giorgii, za zaufanie i wsparcie. To dopiero początek. Zobowiązujemy się do dzielenia się rezultatami naszych prac – będziemy publikować wyniki badań, opisy eksperymentów, materiały edukacyjne oraz wnioski z wykorzystania Transkribusa w projektach naukowych i dydaktycznych. Mam nadzieję, że nasze doświadczenia okażą się wartościowe również dla innych badaczy i instytucji rozwijających Humanistykę Cyfrową.

22/07/2026

Millions of church record pages are available on Archion, but until now, the historical handwriting was often a closed door for anyone who couldn't read Kurrent script.

That is changing.

By integrating the Transkribus API directly into Archion, one of the largest genealogical archives in the German-speaking world, automated text recognition is now available right inside the document viewer.

Users no longer need to switch platforms or export images. Instead, AI-powered transcriptions are generated instantly, right next to the original document.

This intergration allows researchers to navigate and transcribe over 200,000 church books without ever leaving the Archion website.

“Our aim was to bring reliable handwriting recognition directly into the Archion research workflow, giving users practical support exactly where they need it.”- Judith Sutter, CEO, Kirchenbuchportal GmbH

Find out how this integration works in practice and what it changes for users:

Discover how Archion streamlined family history research by integrating AI text recognition through the Transkribus API, making historical documents more accessible.

21/07/2026

How many collections, models, and pages of Ground Truth are you managing right now?
If the answer is "too many," our new Projects and Datasets features can help bring structure to your workflow.

As your projects grow, keeping a clear overview becomes increasingly important. But between collections, AI models, collaborators, and training data, it's easy to lose track of what belongs where.

Datasets help you organise and structure your ground truth, creating traceable training data for custom model training, while Projects bring your collections, models, sites, and datasets together in one central workspace.

If you missed our latest webinar, the recording is now available. We walk through both features, explain how they fit into your workflow, and demonstrate how they can help you to manage your work more effectively.

Watch here:

Managing historical document digitisation projects often means keep...

15/07/2026

What do Irish, Ancient Greek, Latin, German, Dutch, and English have in common?
Historical documents rarely stick to a single language.
It's not unusual to find multiple languages, or even different scripts, on the same page, making automated transcription a real challenge.

Instead of splitting collections by language, researchers are using Transkribus to train custom models that can recognise multiple languages and scripts together.

The University of Galway and New York University, for instance, successfully developed a bilingual model for , a 19th-century newspaper containing both English and Irish text, including traditional Gaelic script.

Our latest blog highlights two more research projects that use multilingual AI models to make complex historical collections more accessible, searchable, and easier to explore.

Discover how researchers are training multilingual models with Transkribus:

Discover how Transkribus is breaking language barriers in historical research with innovative multilingual text recognition projects. Unlock the potential of your archives.

09/07/2026

Were society magazines the social media of the late 19th and early 20th centuries?

The Wiener Salonblatt was the place to see and be seen. With over 300,000 aristocratic notices and 20,000 portrait photographs, it captured the social lives and networks of the Habsburg elite for decades.

Today, Christian Lendl uses Transkribus as the starting point for an AI-powered workflow that transforms this unique collection of text and images into structured data, opening up new ways to explore historical social networks at scale.

See how at TUC 2026 at the Universität Passau: https://eu1.hubs.ly/H0wQgV70

🎤 From Gossip to Structured Data: AI-assisted multimodal extraction from a historical society magazine
📍 23 September · Panel 3: Archives, Research, and Society

Panel 3Scholarship PresentationsWednesday 23 September09:00 – 11:00 Presented by Christian LendlACDH, Austrian Academy of SciencesFrom Gossip to Structured Data: AI-assisted multimodal extraction from a historical society magazine300,000 aristocratic notices and 20,000 portrait photographs in one ...

02/07/2026

We've just released Text Titan II, our most accurate transcription model yet.
As the successor to Text Titan I, it is the next generation of our general-purpose transcription model for printed and handwritten documents across the major Latin-script languages.

Historical collections rarely contain just one type of document. They combine handwriting and print, multiple languages, and centuries of changing scripts.

Finding a single model that performs well across that variety has always been a challenge.

Trained on more than 250 million words, it reduced transcription errors by an average of 47% compared to Text Titan I on our internal benchmark.

The biggest improvements come on handwritten material, helping reduce the time spent correcting transcriptions while making more collections searchable and accessible.

Whether you're digitising an archive, working with historical records or carrying out research, Text Titan II is the best place to start for most Latin-script documents.

👉Read more about Text Titan II and the benchmark results in our blog post: https://eu1.hubs.ly/H0wDXbl0

👉Try out Text Titan II here: https://eu1.hubs.ly/H0wDXcb0