Inference-Only Prompt Projection for Safe Text-to-Image Generation with TV Guarantees

Minhyuk Lee, Hyekyung Yoon, Myungjoo Kang · 2026

Read Inference-Only Prompt Projection for Safe Text-to-Image Generation with TV Guarantees by Minhyuk Lee, Hyekyung Yoon, Myungjoo Kang (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-2602-00616.txt.

Description

Minhyuk Lee and colleagues tackle safe text-to-image generation, suppressing unsafe outputs while keeping benign prompt behaviour intact. They frame the trade-off through total variation: cutting unsafety forces prompt-conditioned deviation from the frozen generator. Their SPOT framework projects prompts toward a tau safe set, using a language model to rank rewrites and a safeguard model to accept images. Across four data sets and three diffusion backbones, SPOT cuts inappropriate scores by up to 44.4% over safety baselines while keeping benign behaviour close to the reference.

Catalog record

CREATORMinhyuk Lee, Hyekyung Yoon, Myungjoo Kang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE84639
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00616

Browse the full library · About the Tower of Babel