Original data, not a screenshot of the page
Inside a PDF, a photo is stored once as an image object, and the page only says where to draw it and how big. A JPEG photo is usually stored as the JPEG file itself, compressed with what the PDF standard calls the DCTDecode filter. This tool reads those objects directly. A JPEG is saved exactly as it is stored, byte for byte, so nothing is lost and nothing is re-compressed. A picture stored with lossless compression is turned into a PNG, the common lossless format, and where its data already uses PNG-style rows it is copied across without being packed again. Each picture keeps the width and height in pixels it has in the file, which is often far larger than it looks on the page: a photo shrunk to a corner of a brochure may be 3000 pixels wide. If the same logo or photo appears on many pages, it is listed once, with all its page numbers.
Which format each picture is saved in
JPEG photos stay JPG. Pictures stored without loss (screenshots, charts, logos, scans saved by many office programs) become PNG. Black-and-white scans made by fax machines and many office scanners use CCITT fax compression; these are saved as TIFF files that carry the original fax data, which Windows Photos, macOS Preview and most image editors open. Pictures stored as JPEG 2000 are saved as .jp2 files, also unchanged, but few programs open them: Safari displayed JPEG 2000 up to version 17, and no major current browser does. JBIG2 is a compression made for scanned text pages; this tool cannot decode it, says so on the card, and you can save those pages as pictures with PDF to JPG instead. Each card shows the planned format before you download.
Transparency, masks and CMYK colours
A PDF keeps transparency next to the picture, as a separate grayscale mask. With "Keep transparency" ticked, the mask is joined back to the picture and the result is saved as a PNG with a real alpha channel, so a logo or cut-out product photo comes out without a white box behind it. Because JPEG cannot hold transparency, a JPEG that has a mask is saved as PNG in that case; untick the option to get the original JPEG instead. Pictures made for print are often stored in CMYK. A CMYK JPEG is saved as it is, which keeps every bit of the original, but some viewers show its colours slightly differently from the PDF. Other CMYK, spot-colour and Lab pictures are converted to ordinary RGB colours when they are saved as PNG.
When you will find fewer pictures than you expect
Not everything that looks like a picture is one. Charts, diagrams, maps and logos drawn by design programs are often vector drawings: lines and shapes, not pixels, so there is no image to pull out. A scanned document is usually one image per page, so you get whole pages back. Some PDFs cut a photo into strips or tiles, which come out as several pieces. Tiny images such as icons, bullets and lines are real images and are listed too; the "Hide tiny images" option removes them from view. When you need each page exactly as it looks, with text and drawings included, use PDF to JPG or PDF to PNG. To get the words from a scanned page, use Image to text; for text PDFs, PDF to text.
Privacy, passwords and large files
The PDF is read by a background thread in this browser tab; it is never sent to a server, and the images are written straight to your downloads. A PDF that needs a password to open asks for it, and the password is used only on your device. Files that only restrict printing or copying open without one. The limit is 300 MB per PDF. Listing is quick even for long files; saving takes longer when there are hundreds of large lossless images, because each one has to be packed into a PNG, and the progress shows which image is being saved. Pictures you take from a PDF may be protected by copyright: use them as the owner of the document allows.