AI-Generated Image Detectors Overrely on Global Artifacts: Evidence from Inpainting Exchange
Elif Nebioglu, Emirhan Bilgiç, Adrian Popescu · 2026
Read AI-Generated Image Detectors Overrely on Global Artifacts: Evidence from Inpainting Exchange by Elif Nebioglu, Emirhan Bilgiç, Adrian Popescu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00192.txt.
Description
Elif Nebioglu and colleagues show that AI image detectors lean too hard on global artifacts from inpainting. Modern inpainting edits local regions with realism, yet detectors often key on side effects spread across the whole image. The authors trace this to VAE reconstruction, which shifts spectra subtly even in untouched regions. Their Inpainting Exchange operation restores pixels outside the edited region to their originals while keeping all made up content. Under that test, top detectors, including commercial ones, fall sharply, at times near chance, from high accuracy.
Catalog record
| CREATOR | Elif Nebioglu, Emirhan Bilgiç, Adrian Popescu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56068 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.00192