AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

PlugMark: A Plug-in Zero-Watermarking Framework for Diffusion Models

Pengzhen Chen, Yanwei Liu, Xiaoyan Gu, Enci Liu, Zhuoyi Shang, Xiangyang Ji, Wu Liu

ICCV 2025 · Image generation › Autoregressive and latent generation · approach: Robust diffusion watermarking · 3 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 image+text input to DKe training instead of image only (ablation)PlugMark · fine-tuned 1000 stepsmatching accuracy (%): 40.67 -> 66.09Improved
generate multimodal (image, text) boundary pairs instead of image-only BRoK (ablation)PlugMark · fine-tuned 1000 stepsmatching accuracy (%): 66.09 -> 85.18Improved
add multi-timestep features to DKe (ablation)PlugMark · fine-tuned 1000 stepsmatching accuracy (%): 85.18 -> 94.82Improved
add uncertainty-based (variance) timestep selection and weighting (ablation)PlugMark · fine-tuned 1000 stepsmatching accuracy (%): 94.82 -> 98.83Improved
add multi-timestep features to DKe (ablation)PlugMark · pruned rho=0.1matching accuracy (%): 90.00 -> 94.90Improved
add uncertainty-based timestep selection (ablation)PlugMark · pruned rho=0.1matching accuracy (%): 94.90 -> 98.00Improved
use PlugMark boundary generation instead of IPguard (ablation)BRoK · fine-tune 2000 stepsTPR: 0.005 -> 0.991Improved
use zero-watermarking PlugMark instead of Stable SignatureStable Diffusion v2 · FIDFID: 24.77 -> 24.25Improved
use PlugMark instead of AquaLoRAStable Diffusion v2 · fine-tune 1000 stepsTPR at FPR=1e-6: 0.003 -> 0.995Improved
use CIFAR-100 (20 classes) instead of CIFAR-10 for DKePlugMark · fine-tuned 20k stepsmatching accuracy (%): 72.55 -> 69.18Hurt
increase number of selected timesteps k (ablation)PlugMark · fine-tuned 1000 stepsmatching accuracyImproved
select timesteps by lowest variance instead of randomly (ablation)PlugMark · fine-tuned 1000 stepsmatching accuracyImproved

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