Pulling The REINS: Training-Free Safety Alignment of Video Diffusion Models via Representation Steering
- URL: http://arxiv.org/abs/2606.17257v1
- Date: Mon, 15 Jun 2026 20:03:09 GMT
- Title: Pulling The REINS: Training-Free Safety Alignment of Video Diffusion Models via Representation Steering
- Abstract summary: We present REINS, a training-free method that aligns video diffusion models at inference time by steering internal representations toward safe generation.<n>Our key finding is that safety-relevant structure is linearly encoded in the hidden-state activations of video diffusion transformers.<n>At inference, adding this direction to hidden states at an intermediate transformer layer redirects generation from harmful content to semantically related safe alternatives, with no weight updates, no concept enumeration, and negligible computational overhead.
- Score: 20.152820762914214
- License: http://creativecommons.org/licenses/by/4.0/
- Abstract: Open-weight video diffusion models can generate photorealistic unsafe content, from violence to misinformation, yet existing defenses either require expensive safety fine-tuning that degrades general capability, or apply external filters that are trivially bypassed by adversarial prompts. We present REINS (REpresentation-space INference-time Safety steering), a training-free method that aligns video diffusion models at inference time by steering their internal representations toward safe generation. Our key finding is that safety-relevant structure is linearly encoded in the hidden-state activations of video diffusion transformers, and a single direction, discovered via Supervised PCA on binary safety labels, suffices to separate safe from unsafe generation trajectories. At inference, adding this direction to hidden states at an intermediate transformer layer redirects generation from harmful content to semantically related safe alternatives, with no weight updates, no concept enumeration, and negligible computational overhead. Through mechanistic analysis, we reveal that while safety information accumulates monotonically with transformer depth, steering effectiveness peaks at intermediate layers (~50% depth), exposing a fundamental tradeoff between information availability and downstream propagation capacity. We evaluate REINS across 9 video diffusion models, multiple parameter scales (1.3B-5B), and both text-to-video and image-to-video generation, to our knowledge, the broadest safety evaluation suite in the video generation literature.
Related papers
- The Price of Consistency: Exploiting Visual Anchors for Multimodal Jailbreaking in Video Generation [83.99528259002504]
We reveal the visual anchoring effect: by enforcing consistency, the mechanism prevents the generated content from drifting away from the original harmful intent.<n>We propose Decoupling Intent via Visual Anchors (DIVA), a training-free multimodal jailbreak framework for video generation that exploits this vulnerability.<n>To facilitate future research, we additionally contribute TI2SafetyBench, the first safety benchmark for multi-conditional video generation.
arXiv Detail & Related papers (2026-09-07T08:34:00Z) - Introspective Attention Modulation for Safe Text-to-Image Generation [37.67434449197168]
State-of-the-art flow based text-to-image (T2I) models exhibit remarkable generative abilities but remain vulnerable to producing unsafe content.<n>We introduce a unique approach that achieves safety by regulating the model's attention dynamics through inference-time introspection.
arXiv Detail & Related papers (2026-07-16T12:53:45Z) - Unified Safe In-context Image Generation in Multimodal Diffusion Transformers via Restricting Unsafe Information Flows [20.386952794426833]
Unified Visual Safety Regulator (UVR) is a training-free safe generation framework that regulates unsafe semantics in generated images.<n>UVR mitigates unsafe generation through unified, targeted attention modulation and explicit restriction of harmful information flow.<n>Experiments show that UVR achieves state-of-the-art safety performance by achieving 91% and 77% erase rate in image synthesis and editing tasks.
arXiv Detail & Related papers (2026-06-05T03:43:04Z) - Do Models Share Safety Representations? Cross-Model Steering for Safe Visual Generation [52.122731171289665]
We introduce the first framework for cross-model safety steering.<n>Our pipeline never accesses unsafe data on the target side.<n>We evaluate our approach in text-to-image and text-to-video generation across diverse source-target model pairs.
arXiv Detail & Related papers (2026-06-03T18:00:04Z) - SafeRedir: Prompt Embedding Redirection for Robust Unlearning in Image Generation Models [67.84174763413178]
We introduce SafeRedir, a lightweight inference-time framework for robust unlearning via prompt embedding redirection.<n>We show that SafeRedir achieves effective unlearning capability, high semantic and perceptual preservation, robust image quality, and enhanced resistance to adversarial attacks.
arXiv Detail & Related papers (2026-01-13T15:01:38Z) - ConceptGuard: Proactive Safety in Text-and-Image-to-Video Generation through Multimodal Risk Detection [27.47621607462884]
ConceptGuard is a framework for proactively detecting and mitigating unsafe semantics in multimodal video generation.<n>A contrastive detection module identifies latent safety risks by projecting fused image-text inputs into a structured concept space.<n>A semantic suppression mechanism steers the generative process away from unsafe concepts by intervening in the prompt's multimodal conditioning.
arXiv Detail & Related papers (2025-11-24T05:27:05Z) - SafeCtrl: Region-Based Safety Control for Text-to-Image Diffusion via Detect-Then-Suppress [48.20360860166279]
We introduce SafeCtrl, a lightweight, non-intrusive plugin that first precisely localizes unsafe content.<n>Instead of performing a hard A-to-B substitution, SafeCtrl then suppresses the harmful semantics, allowing the generative process to naturally and coherently resolve into a safe, context-aware alternative.
arXiv Detail & Related papers (2025-08-16T04:28:52Z) - Video-SafetyBench: A Benchmark for Safety Evaluation of Video LVLMs [45.265397990158846]
Video-SafetyBench is the first benchmark designed to evaluate the safety of LVLMs under video-text attacks.<n>It comprises 2,264 video-text pairs spanning 48 fine-grained unsafe categories.<n>To generate semantically accurate videos for safety evaluation, we design a controllable pipeline that decomposes video semantics into subject images and motion text.
arXiv Detail & Related papers (2025-05-17T05:06:38Z) - Safety Without Semantic Disruptions: Editing-free Safe Image Generation via Context-preserving Dual Latent Reconstruction [88.18235230849554]
Training multimodal generative models on large, uncurated datasets can result in users being exposed to harmful, unsafe and controversial or culturally-inappropriate outputs.<n>We leverage safe embeddings and a modified diffusion process with weighted tunable summation in the latent space to generate safer images.<n>We identify trade-offs between safety and censorship, which presents a necessary perspective in the development of ethical AI models.
arXiv Detail & Related papers (2024-11-21T09:47:13Z) - LLMs know their vulnerabilities: Uncover Safety Gaps through Natural Distribution Shifts [88.96201324719205]
Safety concerns in large language models (LLMs) have gained significant attention due to their exposure to potentially harmful data during pre-training.<n>We identify a new safety vulnerability in LLMs, where seemingly benign prompts, semantically related to harmful content, can bypass safety mechanisms.<n>We introduce a novel attack method, textitActorBreaker, which identifies actors related to toxic prompts within pre-training distribution.
arXiv Detail & Related papers (2024-10-14T16:41:49Z)
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.