OCR engine Azure Document Intelligence — unavailable Calamari — unavailable Google Vision — unavailable Kraken (HTR) — unavailable Mistral OCR — unavailable PERO — unavailable Tesseract Segmenter PP-DocLayout — unavailable Mise en page Tesseract DocLayout-YOLO — unavailable Segmenteur distant (HF) Detects regions at the head of the pipeline; the remote one delegates to an HF endpoint (swap the model via the URL). Per-block recognizer Azure Document Intelligence — unavailable Calamari — unavailable Google Vision — unavailable Kraken (HTR) — unavailable Mistral OCR — unavailable PERO — unavailable Tesseract Anthropic — VLM, zero-shot — unavailable Mistral — VLM, zero-shot — unavailable Ollama — VLM, zero-shot OpenAI — VLM, zero-shot — unavailable Engine applied to each cropped block: a real OCR, or a VLM transcribing the block image (zero-shot). Assembled text scored CER/WER like a flat pipeline. LLM Anthropic — unavailable Mistral — unavailable Ollama OpenAI — unavailable VLM Anthropic — unavailable Mistral — unavailable Ollama OpenAI — unavailable Model Prompt — Curated prompt (by period) — correction_early_modern_english correction_image_medieval_french correction_image_press_19c_english correction_image_press_19c_european correction_image_press_19c_french correction_image_press_19c_german correction_imprime_ancien correction_medieval_english correction_medieval_french correction_press_19c_english correction_press_19c_european correction_press_19c_french correction_press_19c_german zero_shot_imprime_ancien zero_shot_medieval_english zero_shot_medieval_french Pick a curated prompt OR write your own below (free text takes precedence). Extract named entities (NER) — unavailable Adds a NER step at the end of the pipeline; scored (F1) if the corpus carries an entity ground truth. Export ALTO (XML) Tesseract only: produces one ALTO XML per document (geometry + text), downloadable from the report — re-importable into eScriptorium or Transkribus. + Add a competitor