# Toward Trustworthy Portrait Editing: Evaluation of Demographic Misrepresentation in I2I Models

arXiv: https://arxiv.org/abs/2602.16149 · HTML: https://mlsanity.com/paper/2602.16149/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 20 of its 40 references that are in the map.
- First posted on arXiv 18 Feb 2026.

## Builds on (most cited first, 20 of 20)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Qwen-Image Technical Report](https://mlsanity.com/paper/2508.02324/) (2025)
- [Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods](https://mlsanity.com/paper/1804.06876/) (2018)
- [Step1X-Edit: A Practical Framework for General Image Editing](https://mlsanity.com/paper/2504.17761/) (2025)
- [Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scale](https://mlsanity.com/paper/2211.03759/) (2022)
- [OR-Bench: An Over-Refusal Benchmark for Large Language Models](https://mlsanity.com/paper/2405.20947/) (2024)
- [Multi-Reward as Condition for Instruction-based Image Editing](https://mlsanity.com/paper/2411.04713/) (2024)
- [SCoFT: Self-Contrastive Fine-Tuning for Equitable Image Generation](https://mlsanity.com/paper/2401.08053/) (2024)
- [New Job, New Gender? Measuring the Social Bias in Image Generation Models](https://mlsanity.com/paper/2401.00763/) (2024)
- [Perceptions and Realities of Text-to-Image Generation](https://mlsanity.com/paper/2306.08363/) (2023)
- [InstaFace: Identity-Preserving Facial Editing with Single Image Inference](https://mlsanity.com/paper/2502.20577/) (2025)
- [OVERT: A Benchmark for Over-Refusal Evaluation on Text-to-Image Models](https://mlsanity.com/paper/2505.21347/) (2025)
- [Bias Amplification in Stable Diffusion's Representation of Stigma Through Skin Tones and Their Homogeneity](https://mlsanity.com/paper/2508.17465/) (2025)
- [DermDiff: Generative Diffusion Model for Mitigating Racial Biases in Dermatology Diagnosis](https://mlsanity.com/paper/2503.17536/) (2025)
- [Exposing Blindspots: Cultural Bias Evaluation in Generative Image Models](https://mlsanity.com/paper/2510.20042/) (2025)
- [Hidden Bias in the Machine: Stereotypes in Text-to-Image Models](https://mlsanity.com/paper/2506.13780/) (2025)
- [TrueSkin: Towards Fair and Accurate Skin Tone Recognition and Generation](https://mlsanity.com/paper/2509.10980/) (2025)
- [Beyond the Prompt: Gender Bias in Text-to-Image Models, with a Case Study on Hospital Professions](https://mlsanity.com/paper/2510.00045/) (2025)
- [T2IBias: Uncovering Societal Bias Encoded in the Latent Space of Text-to-Image Generative Models](https://mlsanity.com/paper/2511.10089/) (2025)
- [The Digital Mirror: Gender Bias and Occupational Stereotypes in AI-Generated Images](https://mlsanity.com/paper/2510.08628/) (2025)

## Built on by (most cited first, 1 of 1)

- [When the Judge Acts: Auditing VLM-Guided Image Selection on Culturally Situated Prompts](https://mlsanity.com/paper/2610.01243/) (2026)

## Abstract (arXiv)

Instruction-guided image-to-image (I2I) editors are increasingly used in consumer and professional visual workflows, where trustworthiness depends not only on prompt compliance but also on equitable preservation of identity-relevant attributes. We formalize two failure modes: Soft Erasure, where requested edits are weakly realized or silently suppressed, and Stereotype Replacement, where edits introduce unrequested, stereotype-consistent demographic attributes. Using a controlled benchmark of 5,040 edited portraits, we evaluate these failures across three recent open-weight editors with vision-language model scoring and human evaluation. Our results show that identity-preservation failures are pervasive and demographically uneven. In particular, 62--71% of outputs exhibit skin lightening, with Indian and Black source portraits affected at 72--75%, compared with 44% for White source portraits, indicating output-level drift toward lighter or more White-presenting appearances when identity constraints are underspecified. In a mitigation case study, prompt-level appearance constraints reduce race-change scores for non-White source portraits by up to 1.48 points, while leaving White source portraits largely unchanged, without modifying model weights. These findings show that identity preservation is not a uniform property of I2I portrait editing systems, but an unevenly distributed trustworthiness failure with direct social consequences. At deployment scale, such silent distortions can shape AI-mediated self-representation and reinforce representational disparities. We introduce a controlled audit protocol for fairness-aware evaluation and governance of generative editing systems. Project page: https://seochan99.github.io/i2i-demographic-bias