Pdf.Rasterizer

Examples below assume this module is imported with an imports: entry under alias Pdf. Kind references follow that alias — substitute your own if you import it under a different name.

Renders one page of a PDF to image bytes (PNG by default; JPEG or WebP to shrink the payload for a vision model) via pdf.js on a server-side canvas (@napi-rs/canvas). The reported width/height are the rendered pixel dimensions at the configured scale — the coordinate space downstream consumers (Pdf.FormFields, Image.Overlay) measure against.

---

Example

kind: Pdf.Rasterizer
metadata: { name: Render }
scale: 2

- name: page
  inputs:
    data: "${{ steps.fetch.result.bytes }}"
    page: 1
  invoke: !ref Render

---

Configuration

Field	Type	Required	Description
scale	number	no (default 1)	Render scale over the PDF's native size (1 ≈ 72 DPI), capped at 8. Pin to the same value as the Pdf.FormFields consuming the measured coordinates.
format	png | jpeg | webp	no (default png)	Output image format. A per-invocation format takes precedence.
quality	integer	no (default 80)	Encoder quality (1–100) for the lossy formats; ignored for png.

Invocation inputs

Input	Type	Required	Description
data	Uint8Array	yes	Buffered PDF bytes (e.g. the bytes produced by Octet.Decoder, or an S3.Get body collected to bytes).
page	integer	no (default 1)	1-based page number to render.
format	png | jpeg | webp	no	Output image format. Takes precedence over the resource-level format.
quality	integer	no	Encoder quality (1–100) for the lossy formats; ignored for png.

Output

Field	Type	Description
image	Uint8Array	Rendered page as buffered image bytes in the chosen format.
width	integer	Rendered image width in pixels (page width × scale).
height	integer	Rendered image height in pixels (page height × scale).
pageCount	integer	Total number of pages in the document.
scale	number	The render scale the image was produced at — wire it into a Pdf.FormFields invocation's scale so the coordinate contract is a value, not a convention.
mediaType	string	MIME type of the encoded image (image/png, image/jpeg, or image/webp) — wire into a vision message's image part.

Errors

Code	Raised when
ERR_INVALID_INPUT	data is not a Uint8Array, the bytes are not parseable as a PDF, page exceeds the document's page count, format is not png/jpeg/webp, or quality is not an integer between 1 and 100.