Image to Markdown (OCR)

Image to Markdown conversion uses OCR to read printed text from a PNG, JPG, or WebP file and return it as structured Markdown. The accuracy of the output depends heavily on the quality of the source image: high-resolution scans of printed text produce clean results, while low-resolution photos of handwritten notes produce variable output that needs significant manual correction.

This conversion is useful for a specific class of documents: printed materials you do not have in digital form, screenshots of content you need to edit, scanned invoices or forms, physical whiteboards captured in a photo, or any image where the text content matters more than the visual presentation. It bridges the gap between printed content and digital workflows.

The result is plain text with Markdown structure applied where headings and lists were detectable in the source. For most image conversions, the output is closer to raw extracted text than richly formatted Markdown, because the visual hierarchy in an image is harder to interpret than the structured formatting in a native document file. Plan for a review pass to add structure where the OCR produced flat text.

Images lock text in a visual format that cannot be searched, edited, or reused without re-typing. A screenshot of a document, a photo of a whiteboard, or a scanned invoice all contain meaningful text that is inaccessible to search engines, document management systems, and AI tools unless it is converted to a machine-readable format first.

OCR-to-Markdown conversion makes image text accessible to the full range of text-first tooling: you can search the content, include it in documentation, track changes with version control, or feed it to an LLM without copying the text manually. For teams that routinely receive printed documents — signed contracts, physical forms, handout materials — maintaining an OCR workflow reduces the time spent re-entering existing text.

For AI workflows, making image text available as Markdown removes a significant extraction barrier. Language models can process the Markdown directly, while the original image requires a vision-capable model plus additional infrastructure. For documents where the text is all that matters, OCR to Markdown is more efficient than vision-model processing.

Converting image text to Markdown is a lighter-weight alternative to vision model processing when the text content is the only thing you need. Instead of sending an image to a vision-capable API — which requires an API key, incurs per-image costs, and varies in extraction accuracy — OCR produces text that can be directly included in any LLM prompt.

For RAG pipelines that need to make scanned document collections searchable, OCR-to-Markdown conversion is the standard ingestion preprocessing step. The Markdown text can be chunked, embedded, and indexed using the same pipeline as native digital documents, without requiring vision model integration.

For chat workflows where you want to ask ChatGPT or Claude a question about a printed document, converting the image to Markdown first lets you paste the text into the conversation directly. This avoids the image upload step, works with models that do not have vision capability, and gives you the ability to review and correct the OCR output before it influences the model's response.

Image resolution is the single most important factor in OCR accuracy. Low-resolution images — under 150 DPI for scanned documents, or small dimensions for screenshots — produce poor text recognition. Before uploading, try to use the highest-resolution version available. For phone camera photos, move closer to the document and ensure the text is in sharp focus.

Handwriting, watermarks, and text that overlaps with complex backgrounds all degrade OCR accuracy. For handwritten documents, expect significant errors that require manual correction. Mixed documents — partially printed, partially handwritten — produce inconsistent quality across sections. Always review the output before using it in a downstream workflow, especially when numbers, dates, or names are important.

any2markdown uses Microsoft's MarkItDown library for image conversion. MarkItDown supports PNG, JPG, JPEG, and WebP formats. Image OCR in MarkItDown is handled through the optional Azure Document Intelligence integration in the full package, which applies layout analysis alongside OCR to improve reading order reconstruction and table detection.

For best results, use images with high contrast between text and background, clear focus, and minimal image compression artifacts. Printed text on white paper photographed under good lighting consistently produces the most accurate OCR output.