AI Paper Atlas

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

Cert-LAS: Toward Certified Model Ownership Verification for Text-to-Image Diffusion Models via Layer-Adaptive Smoothing

Leyi Qi, Yiming Li, Siyuan Liang, Zhengzhong Tu, Dacheng Tao

ICML 2026 · Image generation › Text-to-image diffusion · approach: Inference-time T2I improvement · 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
add layer-adaptive smoothing (LFS-guided noise) instead of layer-uniform smoothing (ablation)Cert-LAS certified verification · SD v1.4, sigma 1.0VSR: 0.812 -> 1.000Improved
add layer-adaptive smoothing instead of layer-uniform smoothing (ablation)Cert-LAS certified verification · SD v1.4, sigma 1.4certified radius: 0.00 -> 1.428Improved
add inference-time layer-adaptive smoothing (ablation)Cert-LAS · Parameter perturbation, L2 budget 0.2T@10^-6F: 0.000 -> 1.000Improved
add inference-time layer-adaptive smoothing (ablation)Cert-LAS · Full fine-tuning on ArtBench, 2000 stepsT@10^-6F: 0.987 -> 0.991Improved
use trigger-free diffusion-classifier watermark instead of synthetic triggersbackdoor-based T2I watermarking · SD v1.4, prompt-space auditingSin(p): 0.958 -> 0.125Improved
adopt exponential growth schedule for noise samples (ablation)Cert-LAS training · SD v1.4training time: reduces training time by 76.6%Improved
replace the private classifier at verification time with another classifier (ablation)Cert-LAS verification · multiple architecturesverification successHurt
apply adversarial parameter perturbation instead of random perturbation to existing watermarksSleeperMark · SD v1.4 parameter perturbationTPR@FPR=1e-6: 0.009 random vs 0.0003 adversarial magnitude to cause dropHurt
transfer Cert-LAS to SDXL and Sana-1.6BCert-LAS · full fine-tuning 2000 stepsT@10^-6F: stays above 0.988No clear effect
vary class pair, prompt template or number of classes (ablation)Cert-LAS · Table 17T@10^-6F: 1.000 for all configurationsNo clear effect

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