Newest papers

Spatially Gated Demographic Steering for Debiasing Multi-Person Text-to-Image Generation

Venkatesh Thirugnana Sambandham, Torsten Schön · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29651 · cited by 0 works (OpenAlex) · open-access PDF

French

Les modèles texte à image reproduisent les stéréotypes sociaux, et l'atténuation existante applique une correction démographique à l'ensemble de la scène. Dans la génération de scène multi-personnel, cela crée un trade-off: la direction globale améliore l'équilibre agrégé uniquement en homogénéisant les personnes au sein d'une image. Des méthodes basées sur la direction telles que l'arithmétique d'intégration conduisent la diversité intra-image en dessous de la ligne de base non guidée. Les approches de texture et de projection latente déplacent la déviation agrégée avec seulement des gains marginaux dans la diversité intra-image. Nous présentons Spatially Gated Steering (SGS), qui contrôle la démographie de chaque

Machine translation (Meta M2M100); the English original follows.

Abstract

Text-to-Image models reproduce societal stereotypes, and existing mitigation applies one demographic correction to the whole scene. In multi-person scene generation this creates a trade-off: global steering improves aggregate balance only by homogenizing the people within an image. Steering-based methods like Embedding Arithmetic drive the intra-image diversity below the unsteered baseline. Text- and latent-projection approaches shift the aggregate skew with only marginal gains in intra-image diversity. We present Spatially Gated Steering (SGS), which steers the demographics of each person independently by extracting per-person masks from the Diffusion Transformer's (DiT) internal attention maps at a critical denoising timestep and applying a distinct demographic vector within each mask for the remaining steps. SGS needs no fine-tuning, inversion or layout specifications, and uses a lightweight segmentor over frozen DiT attention features to localize subjects at early timesteps. On skewed occupational prompts, SGS reaches high intra-image racial and gender diversity while keeping aggregate balance competitive and image quality near baseline, a result we validate on two architecturally distinct diffusion transformers, FLUX.1-Dev and SD3.5-Large. Codebase and model weights are available at https://github.com/cvims/spatially-gated-steering

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Debiasing, Skew, Computer science, Glyph (data visualization), Aggregate (composite), Offset (computer science)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange