Research
These notes are measured runs on the same ImageProcessor that powers the public tools. They exist so a claim on JPG Smaller can be checked against a table, not against a slogan.
How the dataset is built
Each row is one processLocalPath call. Source files live in the project photo set: alpine, coast, forest, city, ceramic, books, flower, and watch. Those are still photographs and object stills at about 2400 pixels on the long edge. We did not add screenshots, scans, or UI chrome to this first set, so conclusions about those subjects are out of scope.
Original bytes come from the file on disk. Output bytes come from the generated file. Labels are formatted from those integers. The engine is PHP GD via ImageProcessor.
Published experiments
Why these three, and not more
We only published tests we could run on the live engine with the photos we have. A WebP vs AVIF vs JPG bake-off would need matched quality settings and a visual scoring method we have not built. A “best format for websites” guide without those measurements would be a rewrite of other people’s pages. That is why it is not here.
Resize sample (small)
Three photos were also sent through resize to 1080×1080. All three wrote 1080×1080 pixels. That confirms the named box. It is too small a sample to rate visual quality across subjects, so it stays on this index instead of becoming its own article.
| Subject | Original | Output | Pixels in | Pixels out | Change |
|---|---|---|---|---|---|
| Alpine | 503 KB | 134 KB | 2400×1600 | 1080×1080 | 73.4% smaller |
| City | 730 KB | 220 KB | 2400×1600 | 1080×1080 | 69.8% smaller |
| Ceramic | 293 KB | 77.8 KB | 2400×1600 | 1080×1080 | 73.4% smaller |
How to read the tables
“Smaller” and “larger” compare output bytes to the source file. They are not quality scores. A larger file can look closer to the source. A smaller file can still be the correct choice for an upload ceiling. Target tools report whether the ceiling was met, separately from how the picture looks.