Continual erasure method keeps unwanted content from diffusion models
Continual Concept Erasure in Diffusion Models by Suppressing Cross-Edit Interference
Computer Vision and Pattern Recognition
Summary
Sometimes AI image models accidentally generate pictures containing unwanted things like celebrity faces or copyrighted styles. It’s important to remove these unwanted parts without breaking the rest of the model. The authors came up with a way to erase these parts continuously, making sure earlier removals don’t get undone or cause other problems. Their method uses math to carefully block interference between edits, so the model stays useful while respecting new restrictions.
What this means in practice
- •For ai model developers: Remove new unwanted content from pre-trained image generation models over time without damaging their ability to create other images.
- •For digital content moderation teams: Ensure image generation tools comply with copyright and privacy rules by continuously filtering out sensitive or protected concepts.
Authors
Yongliang Wu, Haori Lu, Jinqi Luo, Wei Cao, Xingyu Zhu, Yaoyao Liu
Abstract
Concept erasure removes copyright-protected, privacy-sensitive, or otherwise undesirable concepts from pretrained text-to-image diffusion models to support content governance and compliance. As erasure requests arrive over time, models must remove new targets without undoing prior erasures. Existing methods do not constrain interference across edits: residual perturbations outside the retain set interact and accumulate, degrading unrelated generations and sometimes collapsing previously erased targets into noise. We propose CEASE (Continual Erasure via Adaptive Subspace Editing), a training-free method that imposes two subspace constraints on a closed-form solver. CEASE adds the token representation of the shared replacement to the solver's invariance matrix and, when interference is detected, projects the current update onto the orthogonal complement of dominant output directions extracted from cumulative past updates. A closed-form decomposition attributes the accumulated interference to repeated activation of the shared replacement and overlap between successive update directions, showing that the two constraints suppress these respective sources. Across continual erasure of celebrities, artistic styles, and instances, CEASE achieves the most consistent erase-preserve trade-off, while existing methods either degrade general generation or insufficiently erase targets.