Triggering OCR for images

[goheretech]

I have a question about using this to extract text from images.

I have tested this on every file type I can think of, including youtube links, and am very impressed with the results, but for some reason when I submit .png or .jpg of documents or other text, the return is empty which leads me to believe the ocr isnt triggering.

Is anyone else seeing similar results?

I have converted .pdfs of just images and that was just fine.

[txhno]

Yes, that matches the current behavior.

Plain .jpg / .png does not seem to do built-in OCR by itself. The basic image converter is mostly metadata plus optional LLM image description, so empty output can happen.

If you want actual OCR on images, use Azure Document Intelligence. The OCR plugin is for PDF / DOCX / PPTX / XLSX, not standalone image files.

[VANDRANKI]

markitdown does not run OCR on its own. For plain images (PNG, JPG), it relies on an LLM to describe the image. You can pass an llm_client and llm_model when creating the MarkItDown instance and it will call the model to generate a text description.

If you need actual OCR (extracting the text that appears in the image), the markitdown-ocr plugin handles that. You can install it separately and it hooks in automatically.

So the short answer: built-in support is LLM description, OCR is via plugin.