From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes
Long Ma 1 Zhiyuan Yan 2 Jin Xu 4 Yize Chen 3 Qinglang Guo 1 Zhen Bi 5 6 Yong Liao 1 Hui Lin 7 · 2025
Read From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes by Long Ma 1 Zhiyuan Yan 2 Jin Xu 4 Yize Chen 3 Qinglang Guo 1 Zhen Bi 5 6 Yong Liao 1 Hui Lin 7 (2025) 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-2504-04827.txt.
Description
Long Ma and co-authors seek general artifacts for deepfake face detection in a 2025 paper. Generators vary widely, leaving different traces such as lighting or color mismatches. Teaching a detector every trace is impossible. The team instead hunt for traces that most fakes share.
They sort artifacts into two types. Face inconsistency artifacts arise when complex facial detail clashes with plainer surroundings. Up sampling artifacts are traces left by the generator decoder as it enlarges images. Most fakes show one or both. A data level method creates pseudo fakes to train detectors toward those general signs.
Catalog record
| CREATOR | Long Ma 1 Zhiyuan Yan 2 Jin Xu 4 Yize Chen 3 Qinglang Guo 1 Zhen Bi 5 6 Yong Liao 1 Hui Lin 7 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-04827