Two fundamentally different outputs
Embedded extraction looks for bitmap objects used by a page. Page rendering asks PDF.js to draw the complete page, including text, vectors, masks and compositing. The first may recover a reusable image; the second preserves what a reader sees.
A scan is often one large page bitmap, while a designed brochure may use many small images, masks and vector shapes. There is no single mode that is best for every PDF.
When to extract embedded images
Choose embedded mode when you need photos or scans without page text and borders. Filter tiny dimensions to remove icons. Exact deduplication is useful for repeated logos. The browser may re-encode decoded pixel data to PNG, and the manifest marks the result as re-encoded.
When to render pages
Choose page export for complex masks, vector art, layered content, annotations or when visual fidelity matters more than recovering an internal stream. Higher DPI improves pixel dimensions but increases memory rapidly. Process only the page range you need.