Inference-Time Concept Suppression and Video-Centric Evaluation for Text-to-Video Models

• URL: http://arxiv.org/abs/2607.14194v1
• Date: Wed, 15 Jul 2026 16:26:09 GMT
• Title: Inference-Time Concept Suppression and Video-Centric Evaluation for Text-to-Video Models
• Abstract summary: Text-to-video (T2V) generators can synthesize realistic and temporally coherent videos.<n>Unlike text-to-image concept erasure, T2V unlearning must suppress a target concept that may persist across frames.<n>We propose textbfSIRUS, a training-free inference-time framework for concept-level T2V unlearning.
• Score: 3.414053724037068
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Text-to-video (T2V) generators can synthesize realistic and temporally coherent videos, but controllably removing a target concept from a generator remains difficult. Unlike text-to-image concept erasure, T2V unlearning must suppress a target concept that may persist across frames while preserving non-target subjects, actions, scenes, and temporal structure. We propose \textbf{SIRUS}, a training-free inference-time framework for concept-level T2V unlearning. Given textual aliases of a target concept, SIRUS localizes target-related prompt evidence and suppresses target expression during sampling, without updating the text encoder or denoising network. We further introduce a video-oriented evaluation framework for T2V unlearning that separately measures target forgetting, non-target preservation, video quality, jailbreak robustness, and efficiency, using video-level failure criteria, frame-level residue statistics, paired preservation analysis, VBench-based quality diagnostics, and deployment overhead measurement. Across five safety, object, and style concepts on CogVideoX, SIRUS reaches 70.4\% average forgetting success and 25.7\% average frame hit, compared with 44.4\% / 47.2\% for VideoEraser, while reducing the average VBench quality drop from -0.043 to -0.016, yielding the strongest forgetting-quality trade-off among fully evaluated baselines. Transfer experiments on Wan2.2 further suggest that SIRUS generalizes across modern T2V backbones.

Related papers

• LeVJEPA: Efficient & Scalable Video Pretraining without the Heuristics [50.45297537821907]
  We introduce LeVJEPA, the first video encoder trained under LeJEPA's collapse-free objective.<n>A single encoder is trained with an invariance loss over global and local views of a clip.<n>LeVJEPA matches or surpasses V-JEPA 2 across ViT-S/B/L at 5.6 to 20.8x less pretraining compute.
  arXiv  Detail & Related papers  (2026-08-27T17:26:24Z)
• BRITE: A Benchmark for Reliable and Interpretable T2V Evaluation on Implausible Scenarios [4.853153612241331]
  Existing benchmarks largely overlooked implausible scenarios and do not measure audio-visual alignment.<n>We introduce BRITE, the first framework that unifies implausible prompting, fine-grained assessment of audio-visual consistency, and QA-based interpretable evaluation.
  arXiv  Detail & Related papers  (2026-04-24T21:34:52Z)
• Speculative Decoding for Autoregressive Video Generation [9.370533301840284]
  We introduce SDVG, which brings speculative decoding to block-based autoregressive video diffusion.<n>A 1.3B drafter proposes candidate blocks via four denoising steps; each block is VAE-decoded and scored by ImageReward.<n>Blocks scoring above a fixed threshold tau are accepted into the 14B target's KV cache; the rest are regenerated by the target.
  arXiv  Detail & Related papers  (2026-04-19T12:01:57Z)
• CVA: Context-aware Video-text Alignment for Video Temporal Grounding [16.546053936240362]
  We propose Context-aware Video-text Alignment (CVA), a novel framework to achieve temporally sensitive video-text alignment.<n>Our framework is built on three key components. First, we propose Query-aware Context Diversification (QCD), a new data augmentation strategy that ensures only semantically unrelated content is mixed in.<n>Second, we introduce the Context-invariant Boundary Discrimination (CBD) loss, a contrastive loss that enforces semantic consistency at challenging temporal boundaries.<n>Third, we introduce the Context-enhanced Transformer (CTE), a hierarchical architecture that combines windowed self-attention with bidirectional cross
  arXiv  Detail & Related papers  (2026-03-26T01:53:00Z)
• Shot-Aware Frame Sampling for Video Understanding [10.771430339569795]
  InfoShot is a task-agnostic, shot-aware frame sampler for long-video understanding.<n>Design is guided by an information-theoretic objective that encourages the sampled set to retain high information about both shot structure and sparse within-shot deviations.<n>Experiments show that InfoShot improves anomaly hit rate and downstream Video-QA accuracy under frame number constraints.
  arXiv  Detail & Related papers  (2026-03-18T05:36:57Z)
• STVG-R1: Incentivizing Instance-Level Reasoning and Grounding in Videos via Reinforcement Learning [65.36458157092207]
  In vision-language models (VLMs), misalignment between textual descriptions and visual coordinates often induces hallucinations.<n>We propose a novel visual prompting paradigm that avoids the difficult problem of aligning coordinates across modalities.<n>We introduce STVG-R1, the first reinforcement learning framework for STVG, which employs a task-driven reward to jointly optimize temporal accuracy, spatial consistency, and structural format regularization.
  arXiv  Detail & Related papers  (2026-02-12T08:53:32Z)
• VideoScore2: Think before You Score in Generative Video Evaluation [69.43069741467603]
  VideoScore2 is a multi-dimensional, interpretable, and human-aligned framework that explicitly evaluates visual quality, text-to-video alignment, and physical/common-sense consistency.<n>Our model is trained on a large-scale dataset VideoFeedback2 containing 27,168 human-annotated videos.
  arXiv  Detail & Related papers  (2025-09-26T18:09:03Z)
• ResidualViT for Efficient Temporally Dense Video Encoding [66.57779133786131]
  We make three contributions to reduce the cost of computing features for temporally dense tasks.<n>First, we introduce a vision transformer (ViT) architecture, dubbed ResidualViT, that leverages the large temporal redundancy in videos.<n>Second, we propose a lightweight distillation strategy to approximate the frame-level features of the original foundation model.
  arXiv  Detail & Related papers  (2025-09-16T17:12:23Z)
• T2VShield: Model-Agnostic Jailbreak Defense for Text-to-Video Models [88.63040835652902]
  Text to video models are vulnerable to jailbreak attacks, where specially crafted prompts bypass safety mechanisms and lead to the generation of harmful or unsafe content.<n>We propose T2VShield, a comprehensive and model agnostic defense framework designed to protect text to video models from jailbreak threats.<n>Our method systematically analyzes the input, model, and output stages to identify the limitations of existing defenses.
  arXiv  Detail & Related papers  (2025-04-22T01:18:42Z)
• ReferDINO: Referring Video Object Segmentation with Visual Grounding Foundations [33.74746234704817]
  Video object segmentation (RVOS) aims to segment target objects throughout a video based on a text description.<n>This is challenging as it involves deep vision-level understanding, pixel-level dense prediction andtemporal reasoning.<n>We propose bfReferDINO RVOS that inherits region-level vision-text alignment from foundational visual grounding models.
  arXiv  Detail & Related papers  (2025-01-24T16:24:15Z)
• NSNet: Non-saliency Suppression Sampler for Efficient Video Recognition [89.84188594758588]
  A novel Non-saliency Suppression Network (NSNet) is proposed to suppress the responses of non-salient frames. NSNet achieves the state-of-the-art accuracy-efficiency trade-off and presents a significantly faster (2.44.3x) practical inference speed than state-of-the-art methods.
  arXiv  Detail & Related papers  (2022-07-21T09:41:22Z)

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.