AI Paper Atlas

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

SafeRoPE: Risk-specific Head-wise Embedding Rotation for Safe Generation in Rectified Flow Transformers

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

CVPR 2026 · Image generation › Text-to-image diffusion · approach: Personalized and region-controlled generation · 1 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 SafeRoPE risk-specific head-wise rotationFLUX.1-dev · Nudity, I2PUnsafe rate (%): 10.3 -> 7.0Improved
apply SafeRoPEFLUX.1-dev · Nudity, Unsafe-1KUnsafe rate (%): 38.8 -> 15.4Improved
apply SafeRoPEFLUX.1-dev · Bloody conceptUnsafe rate (%): 68.1 -> 15.5Improved
apply SafeRoPEFLUX.1-dev · COCO-1K utilityVQA score: 87.5 -> 89.7Improved
apply SafeRoPEFLUX.1-dev · COCO-1K utilityCLIP score: 31.3 -> 31.3No clear effect
apply SafeRoPEFLUX.1-dev · COCO-1K utilityFID: 76.8 -> 68.9Improved
transfer rotation matrices learned on FLUX.1-dev to FLUX.1-schnellSafeRoPE · Bloody conceptUnsafe rate (%): 34.2 -> 16.4Improved
transfer dev-trained rotations rather than training on schnell (ablation)SafeRoPE · Bloody concept (FLUX.1-schnell)Unsafe rate (%): 13.7 -> 16.4Hurt
use learned rotations instead of random rotations (ablation)SafeRoPE · Nudity, Unsafe-1KUnsafe rate (%): 35.2 -> 15.4Improved
use independent rotation matrices with scaled image-token initialization instead of shared/unscaled (ablation)SafeRoPE · Nudity, Unsafe-1KUnsafe rate (%): 24.2-29.0 -> 15.4Improved
raise rotation rank from 2 to default (ablation)SafeRoPE · Nudity, Unsafe-1KUnsafe rate (%): 34.0 -> 15.4Improved
raise rotation rank to 10 instead of default (ablation)SafeRoPE · Nudity, I2PUnsafe rate (%): 7.0 -> 11.1Hurt

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