Where the quality goes
A PDF is a container, the same way a folder is a container. Putting an image inside one does not change the image. Every bit of quality loss in a photo to PDF conversion comes from one specific step: a converter decoding your image to pixels and encoding a new JPEG to put in the file.
Whether that step happens depends on your source format, and it is not a matter of opinion:
| Source | Can go into a PDF unchanged | Loss |
|---|---|---|
| JPG, JPEG | Yes, as a DCTDecode stream | None, with the right tool |
| PNG | No | One JPEG encode |
| HEIC, HEIF | No | One decode plus one JPEG encode |
| WebP | No | One decode plus one JPEG encode |
| DNG, raw | No | Uses the embedded preview |
| TIFF | Sometimes, if JPEG compressed inside | Usually one encode |
The first row is the one that matters, because most photos people convert are already JPEGs. That case can be perfect, and most tools throw it away anyway.
The lossless case, and why so few tools do it
The PDF specification has a filter called DCTDecode, which means the stream
that follows is JPEG data. A PDF viewer hands it to a JPEG decoder. It is the same
decoder that would have opened your file directly.
So a correct JPG to PDF converter reads the width, height and component count out of the JPEG's own header, writes a PDF image object with those numbers, and copies the file body across. Nothing is decoded. The conversion is a repackaging.
Most tools skip this because the lazy path is much less code: draw the image onto a canvas, call an encode function, done. It handles every format with one code path, and it silently costs you a generation on the one format where it did not have to.
The converter on this site does the direct copy whenever quality is set to Original and the source is a normal 1 or 3 component JPEG. You can check: a PDF made that way is within a few kilobytes of the sum of the input files.
Generation loss, and why it compounds
JPEG is lossy by design. Encoding discards detail your eye is bad at noticing. Decode and re-encode the same image and the second pass discards a bit more, because it is now working on an image that already has compression artefacts in it, and it treats those artefacts as real detail worth preserving.
One pass at quality 92 is invisible on a photograph. Five passes at quality 92 is visible. Ten is obvious. This is why a photo that has been forwarded through several apps looks soft: each app re-encoded it.
Practical rule: minimise the number of tools your image passes through, and prefer tools that say what they do. If you convert HEIC to JPG, then JPG to PDF, then compress the PDF, that is three passes for a job that needed one.
What to actually set
| Goal | Quality | Page size | Margin |
|---|---|---|---|
| Archive, keep everything | Original | Fit the photo | None |
| Send to a person | Original or High | Fit the photo | None |
| Print at home | Original or High | A4 or Letter | Small |
| Upload form with a size cap | Smaller file | A4 or Letter | None |
| Screenshots and text | Original or High, never lower | Fit the photo | None |
Page size and margin never touch the image data. They decide how large the image is drawn. A photo on an A4 page holds exactly the same pixels as the same photo on a fitted page.
Making it smaller on purpose
Sometimes the requirement is the opposite: a portal caps uploads at 5 MB and you have twelve photos. Three levers, in the order you should pull them.
- Drop the JPEG quality. Going from 92 to 80 typically halves the file and is hard to see on a photograph. This is the cheapest win available.
- Reduce the pixel dimensions. A 12 megapixel photo of a document is far more resolution than any reader needs. Halving each dimension quarters the pixel count.
- Send fewer pages. Obvious, and usually the correct answer once you look at what is actually in the batch.
What not to do: convert at full quality and then run the PDF through a compressor. That is two encodes to get to a size one encode would have reached, and the result is worse.
Checklist
- Source is already JPG: use a tool with a passthrough setting. Loss is zero.
- Source is HEIC, PNG or WebP: one encode is unavoidable. Keep it at high quality and do it once.
- Do not chain converters. Every extra tool is another generation.
- Page size and margins are free. They do not affect the image data.
- Need it smaller: lower the quality on the way in, not by compressing the PDF afterwards.
- Screenshots and text pages punish low quality far more than photographs do.
- Keep the originals. A PDF is a delivery format, not an archive.