A specialist focused on a single task: reading text from images. TrOCR combines a vision encoder with a text decoder, trained in stages to progressively refine its ability to transcribe printed and handwritten content. It operates quietly and precisely, without the broad conversational range of general-purpose models.