PDF to JPG

How it works

A PDF page has no pixels to begin with. It stores its size in points (a point is 1/72 inch), so an A4 page is 595×842 points and a US Letter page is 612×792. Converting to an image means deciding how many pixels each of those inches becomes, which is the only thing the DPI setting does, and the output dimensions follow from it arithmetically. A4 comes out at 595×842 at 72 DPI, 1240×1754 at 150, and 2480×3508 at 300; Letter gives 1275×1650 at 150.

Choosing 150 or 300

Pick 150 unless you have a reason not to: it is legible on any screen and reasonable to email. Pick 300 when the image is going to be printed, since that is roughly what a desktop printer can resolve. Doubling the DPI quadruples the pixel count and quadruples the memory the render needs, so this tool renders, encodes and releases one page before it touches the next.

What rasterising throws away

Rendering flattens the page into a grid of pixels, and everything that made it a document stops existing at that moment. The text is no longer text: you cannot select it, search it, or copy an address out of it, and a screen reader encounters a picture with nothing to read.

Usually that is precisely what you want. People convert a page to an image to drop it into a slide, attach it to a chat message that mangles PDFs, satisfy an upload form that only accepts images, or post a page somewhere that will not render a document inline. In all of those the loss of selectable text costs nothing. If the words are what you actually need, extract them from the PDF directly rather than converting to an image and running OCR over the result, which discards text the file was already carrying and then guesses it back.

PNG against JPEG, measured

The format choice matters more than people expect, and the intuition that JPEG is the small one is right about half the time. JPEG works on 8×8 blocks of frequency coefficients, which suits continuous tone (photographs, photographic scans, gradients) and copes badly with a hard black-on-white edge, scattering faint ringing around every letterform. PNG works on flat colour and runs of identical pixels.

Which one wins turns on how much ink is on the page rather than on whether the page is "text", because the renderer anti-aliases every glyph and a soft edge is a run of unique greys that PNG cannot pack. Measured on real documents at 150 DPI: a sparsely set book page came out 12 KB as PNG against 17 KB as JPEG, and at 300 DPI 29 KB against 48 KB, but a densely typeset one-page résumé went the other way at 212 KB against 167 KB, and a scanned page at 442 KB against 378 KB.

Try PNG first for a page that is mostly white with black marks on it, and take JPEG for scans, photographs and dense pages. The size question is separate from the quality one: PNG is lossless and puts no ringing around a letterform, which is often worth the bytes on a page somebody has to read.

Vector pages versus embedded scans

How much resolution is worth requesting depends on what the page is built from, and the two cases behave completely differently. A vector page is built from embedded fonts, strokes and filled paths, so it has no native resolution: a request for 600 DPI is as legitimate as one for 72, and every increase produces genuinely finer output.

A scanned page is one bitmap placed on the sheet at a fixed pixel size; asking for 300 DPI when the embedded scan was captured at 96 recovers no detail whatsoever, it resamples the same blur across nine times as many pixels and hands you a much larger file for it. Nothing in the PDF announces which sort of page you have, but zooming to 400% in any viewer settles it: crisp edges mean vector, soft ones mean you are looking at a scan and should not pay for DPI you cannot use.

Several pages arrive as a ZIP

When you select more than one page the images come back as a ZIP, written entry by entry so that a fifty-page document at 300 DPI never has to exist in memory at once.