AI Paper Atlas

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

CASteer: Cross-Attention Steering for Controllable Concept Erasure

Tatiana Gaintseva, Andreea-Maria Oncescu, Chengcheng Ma, Ziquan Liu, Martin Benning, Gregory Slabaugh, Jiankang Deng, Ismail Elezi

ICLR 2026 · Image generation › Text-to-image diffusion · approach: Compositional text-to-image guidance · 14 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
add intermediate clipping (ablation)CASteer · I2P nudity removaltotal detected nude parts: 12 -> 7Improved
add intermediate clipping (ablation)CASteer · I2P inappropriate contentoverall inappropriate proportion (%): 28.94 -> 25.58Improved
add intermediate clipping (ablation)CASteer · Remove Van GoghAcc_e: 0.35 -> 0.25Improved
add intermediate clipping (ablation)CASteer · Remove Van GoghLPIPS_e: 0.46 -> 0.44Hurt
add intermediate clipping (ablation)CASteer · Remove Kelly McKernanAcc_e: 0.05 -> 0.05No clear effect
steer cross-attention outputs rather than other layer types (ablation)CASteer · DiT backbone layersconcept erasure effectivenessImproved
use 50 prompt pairs for steering vector (ablation)CASteer · concrete concept erasureconcept captureNo clear effect
use CASteer instead of SD v1.4 unmodifiedSD v1.4 · I2P nuditytotal detected nude parts: 646 -> 7Improved
use CASteer (clip) instead of RecelerReceler · I2P inappropriate contentoverall inappropriate proportion (%): 27.0 -> 25.58Improved
use CASteer instead of SPM for implicit Mickey removalSPM · user studyimages preferred (%): 2.17% -> 84.32%Improved
use CASteer instead of SPM for Snoopy removalSPM · user studyimages containing Snoopy (%): 44.4 -> 3.3Improved
use CASteer instead of SPM for preserving Mickey when removing SnoopySPM · user studyaverage rank (lower = more Mickey kept): 1.05 -> 2.22Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 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