AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

GoodDiffusion: Proactive Copyright Protection for Diffusion Bridge Models via Learnable Sample-specific Signatures

Shixi Qin, Zhiyong Yang, Shilong Bao, Zitai Wang, Qianqian Xu, Qingming Huang

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Robust diffusion watermarking · 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 sample-specific signatures (LSN) instead of static signature (ablation)GoodDiffusion on I2SB · CelebA super-resolutionAbuse Rate under surrogate attack: 98.52 -> 0Improved
add sample-specific signatures instead of static (ablation)GoodDiffusion on I2SB · CelebA inpaintingAbuse Rate under surrogate attack: 69.16 -> 0Improved
add sample-specific signatures instead of static (ablation)GoodDiffusion on I2SB · CelebA deblurringAbuse Rate under surrogate attack: 99.97 -> 0.04Improved
add sample-specific signatures instead of static (ablation)GoodDiffusion on I2SB · CelebA super-resolutionBFID under surrogate attack: 56.97 -> 261.87Improved
protect model with GoodDiffusionDDBM-VP · CelebA super-resolutionFID (authorized): 12.68 -> 16.22Hurt
protect model with GoodDiffusionDDBM-VE · CelebA super-resolutionFID (authorized): 13.42 -> 28.64Hurt
protect model with GoodDiffusionI2SB · CelebA super-resolutionAbuse Rate: 100 -> 0Improved
protect model with GoodDiffusionDBIM · CelebA deblurringPSNR: 45.40 -> 38.49Hurt
raise injection strength gamma from 0.9 to 0.99 (weaker signature) (ablation)GoodDiffusion I2SB · CelebA super-resolutionAbuse Rate: 0 -> 15.88Hurt
lower gamma from 0.9 to 0.1 (stronger signature) (ablation)GoodDiffusion I2SB · CelebA super-resolutionFID (authorized): 28.25 -> 57.38Hurt
fine-tune the protected model for 50 steps (ablation)GoodDiffusion I2SB · CelebA super-resolutionAbuse Rate: 0 -> 100Hurt
fine-tune protected model vs train from scratch, 50 steps (ablation)GoodDiffusion I2SB · CelebA super-resolutionFID: 22.44 -> 75.17Hurt

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