Images

Two image-handling modes, mutually exclusive: write to disk (writeImages) or inline as base64 (embedImages).

Write to disk

await toMarkdown(buf, {
  writeImages: true,
  imagePath: "out/images", // created if missing
  imageFormat: "png",
  dpi: 150,
  filename: "paper.pdf", // affects the generated filename prefix
});

Files are written as <imagePath>/<filename-basename>-<pageIndex>-<imageIndex>.<ext> and referenced from the Markdown as ![image-N-I](path).

Inline as base64

await toMarkdown(buf, { embedImages: true });

Each image becomes a ![image-N-I](data:image/png;base64,…) link. Useful when downstream consumers (LLMs, blob stores) can ingest the markdown without filesystem access. Watch the size — embedding all images in a PDF can blow up the markdown buffer by 10×+.

Size threshold

imageSizeLimit (default 0.05) skips images smaller than this fraction of the corresponding page edge. Bump it to 0.1 if you want only large figures, drop to 0.01 to keep everything.

await toMarkdown(buf, { embedImages: true, imageSizeLimit: 0.1 });

DPI

dpi (default 150) controls how the bounding region is rasterized. Lower = smaller files. Higher = sharper but bigger.

Format

imageFormat: "png" | "jpg" | "jpeg". JPEG uses quality 85.

Inspecting

const chunks = await toMarkdownPages(buf, { writeImages: true, imagePath: "out" });
for (const c of chunks) {
  for (const img of c.images) {
    console.log(img.bbox, img.ref); // ref is the saved file path
  }
}

Caveats

The bounding region is rasterized via page.toPixmap — what gets embedded is the page region, not the original image stream. This means surrounding text or vector overlays in that region also end up in the snapshot. Pass a tighter imageSizeLimit to skip incidental glyphs that happen to live next to figures.