How the signals work
- Noise (experimental): Immerkær's fast noise estimate on the flattest detailed area. On our real-photo test set it found about 4 in 10 visibly noisy images, with about 7% false alarms.
- JPEG compression (experimental): the encoder quality recorded in the file plus an 8×8 blocking measure. Too few heavily compressed real photos were available to judge it properly.
What we switched off, and why
We also built automatic checks for focus, exposure and blown highlights. They worked on controlled test images but, on 96 real photographs from phones, cameras and AI tools, they missed most of the problem images (focus: 0 of 26). Phone sharpening, film grain and JPEG noise fool simple blur measures. Rather than show a reassuring result we cannot back up, these checks are off. Look at your photos at 100% zoom for these.
These signals are hints, not verdicts, and they are our own heuristics — not Adobe's thresholds. Banding and pixelation are not checked.
Submitting to Adobe Stock? The Adobe Stock Submission Checker runs these signals together with Adobe's published technical requirements and near-duplicate detection.