Auto

OCR runs, with or without VLM augmentation. With the VLM on, OCR reads the whole page and the VLM corrects and completes that read; the VLM alone reads only the regions the layout detector kept, so text outside them, and table content, would otherwise be missing. This is the default.