AI RESEARCH
Inference-Only Prompt Projection for Safe Text-to-Image Generation with TV Guarantees
arXiv CS.AI
•
ArXi:2602.00616v2 Announce Type: replace Text-to-Image (T2I) diffusion models enable high quality open ended synthesis, but practical use requires suppressing unsafe generations while preserving behavior on benign prompts. We study this tension relative to the frozen generator, using its prompt conditioned distribution as the preservation reference. Since T2I safety is commonly evaluated by bounded risk scores on generated images, total variation (TV) bounds how much expected risk can change from this reference.