AI Paper Atlas

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

Localized Concept Erasure in Text-to-Image Diffusion Models via High-Level Representation Misdirection

Uichan Lee, Jeonghyeon Kim, Sangheum Hwang

ICLR 2026 · Image generation › Text-to-image diffusion · approach: Concept erasure in diffusion models · 2 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 misdirection loss to first-block output (early layer) instead of final block (ablation)Diff-Q* (early layer random misdirection) · I2P-931 / COCO-1kLPIPS, CLIP: CLIP 0.273 -> 0.187; I2P 7.09% -> 0.75%Hurt
supervise with final-block high-level representation while updating first block (ablation)HiRM-R · text encoder erasureretention vs erasureImproved
use final-layer W_out as target rather than early-layer targets (ablation)HiRM-R · target-layer ablationerasure and retentionImproved
replace random misdirection with semantic misdirection (ablation)HiRM-R -> HiRM-S · UnlearnCanvas object erasingAA: 94.65 -> 96.94Improved
replace random misdirection with semantic misdirection (ablation)HiRM-R -> HiRM-S · UnlearnCanvas style erasingAA: 94.24 -> 95.54Improved
replace random misdirection with safety vector (ablation)HiRM-R -> HiRM-S · I2PASR (%): 0.96 -> 0.66Improved
replace random misdirection with safety vector (ablation)HiRM-R -> HiRM-S · COCO-10kFID: 8.05 -> 6.75Improved
add HiRM-R text encoder to ESDESD on Flux1.dev · Ring-16ASR (%): 41.05 -> 12.63Improved
add HiRM-R text encoder to EraseAnythingEraseAnything on Flux1.dev · Ring-16ASR (%): 29.47 -> 3.16Improved
add HiRM-R text encoder to CACA on Flux1.dev · MMAASR (%): 3.20 -> 4.40Hurt
add HiRM-S text encoder to SPEED U-NetSPEED · 50-celebrity retentionAcc_r: 88.48 -> 79.30Hurt
merge six style LoRA modules by plain averagingHiRM · UnlearnCanvas multi-stylein-domain accuracy: UA 88.33; IRA 65.56; CRA 98.8Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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