EraseSAE: Surgical Concept Erasure in Text-to-Video Diffusion Models via Sparse Autoencoders

• URL: http://arxiv.org/abs/2609.03629v2
• Date: Tue, 08 Sep 2026 07:39:53 GMT
• Title: EraseSAE: Surgical Concept Erasure in Text-to-Video Diffusion Models via Sparse Autoencoders
• Abstract summary: Concept erasure offers a principled remedy by removing unwanted semantics from pretrained models.<n>Existing approaches typically operate at a coarse quality, misaligned with the finegrained, distributed nature of concept representations.<n>We propose EraseSAE, a novel framework that leverages sparse autoencoders to achieve surgical concept erasure.
• Score: 103.46709005379591
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Recent advances in text-to-video (T2V) diffusion models have demonstrated remarkable generative capabilities, yet their reliance on loosely curated training data raises pressing safety and copyright concerns. Concept erasure offers a principled remedy by removing unwanted semantics from pretrained models while preserving remaining concepts. However, existing approaches typically operate at a coarse granularity misaligned with the fine-grained, distributed nature of concept representations, leading to incomplete removal or degraded generation quality. We argue that surgical erasure fundamentally requires intervention at the level of monosemantic features, where each unit encodes a single interpretable concept. To this end, we propose EraseSAE, a novel framework that leverages sparse autoencoders to achieve surgical concept erasure in DiT-based T2V diffusion models via a principled decompose-attribute-erase pipeline. We first introduce the Partitioned Convolutional Sparse Autoencoder, which decomposes dense spatiotemporal activations into disentangled, interpretable sparse features while preserving spatiotemporal coherence. A contrastive attribution mechanism then contrasts activations from paired prompts to isolate concept-specific feature kernels. At inference, timestep-resolved spatiotemporal masks derived from the identified kernels confine erasure to regions where the target concept is active, leaving unrelated content intact. Extensive experiments across diverse diffusion models and concept erasure tasks demonstrate that EraseSAE achieves precise and robust concept removal with minimal quality degradation, substantially outperforming state-of-the-art methods. The code is available at https://github.com/HiDream-ai/EraseSAE.

Related papers

• Interpreting Object-Dependent Concept Brittleness in Text-to-Image Diffusion Models [105.53190946111586]
  We term this phenomenon object-dependent concept brittleness.<n>We present an interpretability-oriented framework to audit and minimally correct these failures.<n>We evaluate the proposed framework on style and attribute failure cases across multiple diffusion backbones.
  arXiv  Detail & Related papers  (2026-09-09T09:07:40Z)
• PEAK: Precise and Persistent Concept Erasure via k-Sparse Autoencoders [25.995968388013555]
  Erasing concepts from large-scale text-to-image (T2I) diffusion models has become increasingly crucial due to the growing concerns over copyright infringement, privacy violations, and offensive content.<n>We propose PEAK, a textbftextitprecise and textbftextitpersistent concept erasure framework via k-Sparse Autoencoders (kSAEs)
  arXiv  Detail & Related papers  (2026-08-11T14:38:01Z)
• Uni-AdaVD: Universal Concept Erasure for Visual Generation via Orthogonal Value Decomposition [62.947473058878565]
  We present Uni-AdaVD, a universal inference-time concept erasure framework for visual generation.<n>Our results suggest that Uni-AdaVD provides an efficient and adaptable safety mechanism for modern visual generative models.
  arXiv  Detail & Related papers  (2026-07-16T03:16:51Z)
• Differential Vector Erasure: Unified Training-Free Concept Erasure for Flow Matching Models [49.10620605347065]
  We propose Differential Vector Erasure (DVE), a training-free concept erasure method specifically designed for flow matching models.<n>Our key insight is that semantic concepts are implicitly encoded in the directional structure of the velocity field governing the generative flow.<n>During inference, DVE selectively removes concept-specific components by projecting the velocity field onto the differential direction, enabling precise concept suppression without affecting irrelevant semantics.
  arXiv  Detail & Related papers  (2026-02-01T08:05:45Z)
• Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models [27.672305802461377]
  We introduce a novel training-free, zero-shot framework for concept erasure that operates directly on text embeddings before the diffusion process.<n>We achieve superior completeness and robustness while preserving locality and image quality.<n>This robustness also allows our framework to function as a built-in threat detection system, offering a practical solution for safer text-to-image generation.
  arXiv  Detail & Related papers  (2025-10-26T22:04:17Z)
• TRACE: Trajectory-Constrained Concept Erasure in Diffusion Models [0.0]
  Concept erasure aims to remove or suppress specific concept information in a generative model.<n>Trajectory-Constrained Attentional Concept Erasure (TRACE) is a novel method to erase targeted concepts from diffusion models.<n>TRACE achieves state-of-the-art performance, outperforming recent methods such as ANT, EraseAnything, and MACE in terms of removal efficacy and output quality.
  arXiv  Detail & Related papers  (2025-05-29T10:15:22Z)
• CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Models [7.68494752148263]
  CURE is a training-free concept unlearning framework that operates directly in the weight space of pre-trained diffusion models.<n>The Spectral Eraser identifies and isolates features unique to the undesired concept while preserving safe attributes.<n>CURE achieves a more efficient and thorough removal for targeted artistic styles, objects, identities, or explicit content.
  arXiv  Detail & Related papers  (2025-05-19T03:53:06Z)
• Sparse Autoencoder as a Zero-Shot Classifier for Concept Erasing in Text-to-Image Diffusion Models [17.694838443796026]
  Interpret then Deactivate (ItD) is a novel framework to enable precise concept removal in T2I diffusion models.<n>ItD uses a sparse autoencoder to interpret each concept as a combination of multiple features.<n>It can be easily extended to erase multiple concepts without requiring further training.
  arXiv  Detail & Related papers  (2025-03-12T14:46:40Z)
• TRCE: Towards Reliable Malicious Concept Erasure in Text-to-Image Diffusion Models [53.937498564603054]
  Recent advances in text-to-image diffusion models enable photorealistic image generation, but they also risk producing malicious content, such as NSFW images.<n>To mitigate risk, concept erasure methods are studied to facilitate the model to unlearn specific concepts.<n>We propose TRCE, using a two-stage concept erasure strategy to achieve an effective trade-off between reliable erasure and knowledge preservation.
  arXiv  Detail & Related papers  (2025-03-10T14:37:53Z)
• Reliable and Efficient Concept Erasure of Text-to-Image Diffusion Models [76.39651111467832]
  We introduce Reliable and Efficient Concept Erasure (RECE), a novel approach that modifies the model in 3 seconds without necessitating additional fine-tuning. To mitigate inappropriate content potentially represented by derived embeddings, RECE aligns them with harmless concepts in cross-attention layers. The derivation and erasure of new representation embeddings are conducted iteratively to achieve a thorough erasure of inappropriate concepts.
  arXiv  Detail & Related papers  (2024-07-17T08:04:28Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.