PDF to JSON: Image Handling Options and Heading Detection

The PDF to JSON converter now gives you control over what happens to the pictures in a document and marks heading-like paragraphs for you. The new Images parameter can keep just a positioned marker, embed each image as a base64 data URI, or turn images into AI generated text so nothing in the document stays invisible to your model. Heading detection adds a headingLevel field to paragraphs that look like headings, giving retrieval pipelines natural chunk boundaries.

Highlights

• Choose how images appear in the JSON output: a positioned marker, embedded base64 content, or AI generated text.
• In describe mode, scanned text and tables are transcribed and charts and photos become short descriptions, so image content is searchable in LLM and RAG scenarios.
• Paragraphs that look like headings carry a headingLevel field, detected by font size relative to the body text.
• Image alt text is preserved in the output.

Parameters

• Images (collection, default remove) - controls how images appear in the JSON output. remove keeps only the image marker with its position, embed adds each image's content as a base64 data URI, and describe replaces the content with AI generated text.
• DetectHeadings (bool, default true) - mark paragraphs that look like headings with a headingLevel field.

Try it out

• PDF to JSON - API sandbox: pdf-to-json