Evaluating Demographic Misrepresentation in Image-to-Image Portrait Editing
Huichan Seo, Minki Hong, Sieun Choi, Jihie Kim, Jean Oh, [0.3ex] Carnegie Mellon University Dongguk University Lavoro AI Research[0.3ex] · 2026
Read Evaluating Demographic Misrepresentation in Image-to-Image Portrait Editing by Huichan Seo, Minki Hong, Sieun Choi, Jihie Kim, Jean Oh, [0.3ex] Carnegie Mellon University Dongguk University Lavoro AI Research[0.3ex] (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-2602-16149.txt.
Description
Huichan Seo leads a team of five researchers auditing image-to-image editors for demographic misrepresentation. Trust in such editors rests not only on following prompts. It also rests on preserving identity-relevant traits fairly. The authors formalize two failure modes. Soft Erasure means requested edits are weakly realized or silently suppressed. Stereotype Replacement means edits add unrequested traits that fit a stereotype.
Using a controlled benchmark of thousands of edited portraits, they evaluate three recent open-weight editors. Scoring combines vision-language models with human review. Failures of identity preservation prove pervasive and uneven. A large majority of outputs show skin lightening. Indian and Black source portraits are affected far more often than White source portraits. In a mitigation test, prompt-level appearance constraints cut race-change scores for non-White sources, while leaving White sources largely unchanged, without changing model weights.
Catalog record
| CREATOR | Huichan Seo, Minki Hong, Sieun Choi, Jihie Kim, Jean Oh, [0.3ex] Carnegie Mellon University Dongguk University Lavoro AI Research[0.3ex] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-16149