arXiv cs.AI· Avishree Khare, Hideki Okamoto, Bardh Hoxha, Georgios Fainekos, Rajeev Alur·· 2 天前AI 评分27
用时序逻辑规范引导文本到视频生成：TSG 框架提升时序约束满足度
Temporal Specification Guidance for Text-to-Video Generation
AI 导读
研究者提出 Temporal Specification Guidance（TSG），一种免训练的推理时框架，将时序逻辑规范转化为可执行目标来引导视频生成采样。该框架用可微定量语义对视觉概念证据轨迹打分，并以梯度构建规范引导的去噪方向，增强 CFG。在覆盖 11 类时序约束的 220 条提示词基准上，该方法在五个模型上将规范满足准确率较 CFG 提升 9.5-20 个百分点，且内存开销极小。
正文 · 原文
Abstract:Text-to-video models often generate the visual concepts requested in natural language prompts but fail to satisfy their intended ordering, timing, and persistence requirements. We propose representing these requirements as specifications in a formal Temporal Logic, and introduce Temporal Specification Guidance (TSG), a training-free inference-time framework that turns temporal specifications into executable objectives for guiding video generation during sampling. TSG scores traces of visual concept evidence using differentiable quantitative semantics and uses the score's gradients to construct a specification-guided denoising direction. We instantiate this framework with video-text attention traces, efficient differentiable scores for Metric Temporal Logic, and gradient-derived attention biases. This instantiation requires neither an external concept classifier nor backpropagation through the video transformer. Its attention-steered forward pass yields a denoising direction that augments classifier-free guidance (CFG). On a benchmark of 220 prompts spanning 11 categories of temporal constraints, including ordered object entry and state transitions, this instantiation improves specification satisfaction accuracy over CFG by 9.5-20 percentage points across five models, with minimal memory overhead. These results demonstrate the potential of temporal specifications and differentiable satisfaction objectives as a flexible interface for controlling video generation.
| Subjects: | Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2610.20822 [cs.AI] |
| (or arXiv:2610.20822v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.20822 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Avishree Khare [view email]
[v1]
Wed, 14 Oct 2026 19:09:33 UTC (23,419 KB)
来源：arXiv cs.AI · arxiv.org