AI historical transcription: Transkribus vs eScriptorium vs FromThePage

Turning manuscript images into useful text

Historical handwriting transcription is more than extracting text from a modern form. A project may need to preserve line layout, adapt to a specific hand, or coordinate many reviewers. Transkribus, eScriptorium, and FromThePage offer different balances of automatic recognition, model control, and collaboration for researchers and archives working from page images.

Transkribus

Transkribus combines hosted AI text recognition with an editor, document layout, and custom model training. Researchers can choose recognition models for their material, correct the resulting text against the image, and build training data for a particular script or handwriting collection. Its plan structure controls processing credits, storage, seats, and access to advanced tools. The current Free tier provides fifty credits each month, making it a practical entry point for a small collection. Scholar adds advanced recognition tools and priority processing, with an annual card showing nine hundred credits and one seat. Current live pricing also lists API access for Scholar and above, at half the app credit rate. The important operational distinction is processing consumption: handwriting recognition, line detection, and table or field steps can affect the credit total differently. Transkribus suits researchers who want recognition and correction in a managed environment without operating their own server. Model suitability still depends on the actual documents and transcription conventions.

eScriptorium

eScriptorium provides an open-source workspace for the full historical transcription process, using Kraken for segmentation and recognition. A team can import document images, identify regions and lines, run trained models, correct the text, and train or fine-tune models for a particular hand or layout. Image-linked editing and support for complex scripts help researchers keep corrections tied to the source page. Exports include structured formats such as ALTO and PAGE XML, and trained models can be downloaded for reuse in the Kraken ecosystem. That portability is a substantial benefit when the model itself is a research output or when an institution needs control of its infrastructure. eScriptorium is decentralized: the core team does not provide an unrestricted public instance, and university-hosted instances apply their own access rules. It therefore suits an institution with hosting capability or access to a suitable research instance. Open-source access removes a core software subscription, while computing, storage, administration, and external support still require resources.

FromThePage

FromThePage organizes transcription around archival projects and the people reviewing them. Institutions host page images; volunteers or staff transcribe, translate, index, and correct the material together. Optional AI Drafts now use Gemini models to create starting text within that workflow, including field-based projects. A page-level AI tab records information such as token use and error measures where human-corrected text is available, keeping the draft’s origin inspectable during review. A separate Transkribus integration offers AI Assist and carries per-page charges. This makes FromThePage particularly useful when an archive’s main challenge is coordinating a community or staff team around AI-produced drafts. Current pricing lists Gemini handwriting recognition as optional on annual Small Organization and Large Institution plans. Free volunteer accounts and the trial do not mean free institutional AI hosting. The platform is therefore best compared with the other products as a collaborative production environment, rather than as a standalone recognition engine with identical model-training controls.

Recognition control versus review workflow

Transkribus combines managed recognition, correction, and custom models in one hosted service. eScriptorium gives research teams greater responsibility for deployment and direct portability of trained Kraken models. FromThePage puts project organization, transcription conventions, volunteer participation, and review around optional AI drafts. For a single researcher, Transkribus offers a comparatively direct starting route. For an institutional model-development project, eScriptorium’s reusable models are compelling. For an archive organizing many human contributors, FromThePage’s collaboration is central to the product choice.

Preserving and reusing the work

eScriptorium explicitly supports downloading trained models and exporting page geometry with text in standard formats. Transkribus offers export and collaboration features according to plan, making the intended publication and team size relevant to selection. FromThePage supports TEI, CSV, HTML, and IIIF/Open Annotation exports and keeps completed inactive projects readable and exportable. Its separate Transkribus connection can also connect human-corrected material to training workflows. These differences matter when the desired result includes reusable models, structured archive records, or an ongoing public transcription project.

What the current prices actually cover

Transkribus Scholar shows €99 per year, equivalent to €8.25 per month with annual billing and displayed Austrian VAT. Its Team annual card and FAQ disagree, so that price needs confirmation. eScriptorium has no core subscription charge, with hosting and compute separate. FromThePage Small Organization lists $3,600 annually or $360 monthly, but optional Gemini HTR is annual-only; its exact usage entitlement needs confirmation. The cheaper Researcher hosting tier does not list that HTR option. Compare processing credits, AI usage, hosting, and reviewer capacity separately within the actual chosen plan.

Which should you choose?

Use Transkribus for a hosted handwriting-recognition and correction workflow, eScriptorium for institutional control over training and reusable models, or FromThePage for AI drafts reviewed within collaborative archival projects. The best choice follows the work you need to manage after recognition: detailed correction, model development, or coordinated human contribution.