Two things get called "getting the images out of a PDF" and they produce completely different files. Knowing which one you want takes ten seconds and saves a lot of frustration.
A picture of the page, or the picture on the page
Converting a PDF to JPG or PNG renders each whole page — text, layout, margins and all — into an image at a resolution you choose. You get a photograph of the document.
Extracting embedded images does the opposite. It reaches inside the file and hands you the individual pictures stored in it, at the size and quality they were stored at, with nothing re-rendered. A page containing a product photo and three paragraphs yields exactly one file: the product photo.
If you want to show someone what the page looks like, render it. If you want the photograph that was placed on it, extract it. The extracted version is almost always higher resolution than a page render, because it is the original asset rather than a rasterisation of the page it sits on.
Why JPEGs come out with no loss at all
This is the part that surprises people. A JPEG placed into a PDF is usually stored as a JPEG — the compressed bytes are embedded more or less intact. Extracting it can therefore be a byte-for-byte copy: nothing is decoded, nothing is re-encoded, and nothing degrades.
That is meaningfully better than any workflow involving a screenshot or a page render, both of which re-compress. Filum hands back those exact bytes. Images stored in other ways are rebuilt as lossless PNGs from the data in the file.
The images that cannot be extracted, and why saying so matters
PDF can store image data in a number of encodings, and some are considerably harder to decode than others: JPEG 2000, palette-indexed colour, 16-bit channels, and images saved with a predictor or a separate transparency mask.
A tool has two options for those. It can guess, and hand you a picture with wrong colours or a broken alpha channel that looks plausible enough to use before you notice. Or it can skip them and say exactly how many it skipped.
Filum does the second. You are told how many images were extracted and how many were skipped, so a partial result is visible as a partial result. Being quietly given fewer images than the document contains is the failure mode that costs the most, because there is nothing in the output to indicate anything is missing.
What comes back
One image downloads on its own. Several arrive as a .zip named after your PDF, one file per image, JPEGs keeping .jpg and PNGs keeping .png so they open anywhere. An image used on several pages is saved once rather than once per page — a logo in a header should not produce forty identical files.
If the PDF contains no embedded images, you are told plainly and nothing downloads. A document whose pages are entirely text has no images to extract: the words you see are drawn from fonts, not stored as pictures. No empty file, no empty archive.
The PDF is opened and the images pulled out inside your browser, on your own device. The file never leaves your computer.