AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Unified Safe In-context Image Generation in Multimodal Diffusion Transformers via Restricting Unsafe Information Flows

Xiang Yang, Feifei Li, Mi Zhang, Geng Hong, Xiaoyu You, Mi Wen, Min Yang

ICML 2026 · Image generation › Text-to-image diffusion · approach: Concept erasure in diffusion models · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply UVR localized attention regulation instead of existing erasure baselinesFLUX.1-dev · I2P (sexual) T2Iunsafe images detected: 207 -> 40Improved
apply UVR to adversarial Unsafe-1K promptsFLUX.1-dev · Unsafe-1K T2Iunsafe images detected: 449 -> 97Improved
apply UVR to editing with unsafe reference imagesFLUX.1-kontext · I2P I2I editingunsafe images detected: 175 -> 46Improved
apply UVR to editing: effect on VQA of unsafe-reference editsFLUX.1-kontext · I2I editingVQA (%): 81.79 -> 88.42Improved
apply UVR to T2I generationFLUX.1-dev · MS-COCOCLIP score: 31.31 -> 31.32No clear effect
add spatial connectivity refinement of unsafe masks (ablation)UVR T2I · Nude T2Iharm (%): 19.9 -> 9.3Improved
add spatial connectivity refinement of unsafe masks (ablation)UVR I2I · Nude I2I editingharm (%): 66.7 -> 25.1Improved
add feature noise perturbation in the start-up stage (ablation)UVR T2I · Nude T2Iharm (%): 39.4 -> 9.3Improved
add feature noise perturbation in the start-up stage (ablation)UVR T2I · Pikachu T2Iharm (%): 100 -> 27.9Improved
apply perturbation in all diffusion steps instead of only the start-up stage (ablation)UVR T2I · Nude and Weapon T2Iharm and CLIP: Nude harm 9.3 -> 0.8; Weapon CLIP 30.92 -> 30.02Mixed
add attention attenuation toward text tokens for I2I (ablation)UVR I2I · Nude I2I editingharm (%): 6.9 -> 25.1 harm, CLIP(u) 22.53 -> 24.21Mixed
add enhanced Oimg-Oimg self-attention (ablation)UVR I2I · Nude I2I editingharm (%): 34.7 -> 25.1Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy