@docture/raster-canvas

NapiCanvasRasterizer renders PDF pages to PNG or JPEG, using pdfjs-dist to draw onto a @napi-rs/canvas surface.

pnpm add @docture/raster-canvas

import { withRasterizer } from "@docture/core";
import { DocumentLoaderPdfJs } from "@docture/loader-pdfjs";
import { NapiCanvasRasterizer } from "@docture/raster-canvas";

// A loader that both reads text AND renders pages. The type says so.
const loader = withRasterizer(new DocumentLoaderPdfJs(), new NapiCanvasRasterizer({ dpi: 300 }));

Rendering is its own plugin type so that "which library renders pages" and "which library reads them" stay independent decisions. A MuPDF renderer can feed a tesseract.js OCR pass.

Options

Set defaults on the constructor; override per call.

| Option | Default | | |---|---|---| | dpi | 300 | 300 for OCR, ~200 is plenty for a vision model | | format | "png" | "png" or "jpeg" | | quality | 0.9 | JPEG only, 0 to 1 | | maxEdge | none | cap the long edge in pixels, and it wins over dpi when they conflict | | pages | all | 1-based page numbers |

RasterPage.scale is points per pixel, computed from what the rasterizer actually produced, so it stays correct after a maxEdge downscale, and pixel boxes can be mapped back to PDF points.

Both libraries load lazily at first render, so constructing the rasterizer is free.