Research map / Video / Video generation
TempoControl: Temporal Attention Guidance for Text-to-Video Models
Shira Schiber, Ofir Lindenbaum, Idan Schwartz
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply TempoControl attention steering to Wan2.1 1.3B | Wan2.1-S · One-object temporal control | Temporal Accuracy: 63.94 -> 83.56 | Improved |
| apply TempoControl attention steering to Wan2.1 1.3B | Wan2.1-S · Two-object temporal control | Temporal Accuracy: 37.50 -> 53.17 | Improved |
| apply TempoControl attention steering to Wan2.1 1.3B | Wan2.1-S · Movement temporal control | Temporal Accuracy: 19.00 -> 54.00 | Improved |
| apply TempoControl to CogX-S | CogX-S 2B · One-object temporal control | Temporal Accuracy: 58.15 -> 63.75 | Improved |
| add explicit timing cues to the text prompt | Wan2.1 · One-object temporal control | Temporal Accuracy: 65.18% -> 63.94% | No clear effect |
| add explicit timing cues to the text prompt | Wan2.1 · One-object temporal control | Imaging Quality: 59.99% -> 53.76% | Hurt |
| add Pearson correlation loss alone (ablation) | TempoControl (Wan2.1-S) · One-object temporal control | Temporal Accuracy / Imaging Quality: Acc 63.94% -> 81.19%; Img 53.76% -> 50.96% | Mixed |
| add entropy regularization alone (ablation) | TempoControl (Wan2.1-S) · One-object temporal control | Imaging Quality: 53.76% -> 59.52% | Improved |
| add entropy to correlation loss (ablation) | TempoControl (Wan2.1-S) · One-object temporal control | Imaging Quality: 50.96% -> 57.60% | Improved |
| add magnitude term to correlation plus entropy (ablation) | TempoControl (Wan2.1-S) · One-object temporal control | Temporal Accuracy: 78.38% -> 82.50% | Improved |
| remove explicit timing cues from prompt for Wan2.2 | Wan2.2 5B · One-object temporal control | Imaging Quality / absence accuracy: Img 48.13% -> 57.78%; Abs 87.42 -> 71.25% | Mixed |
| apply TempoControl on VBench multiple objects | Wan2.1-S · VBench Multiple Objects | Multiple Object (GRiT): 74.13% -> 76.37% | Improved |
Closest papers by meaning
- VSTAR: Generative Temporal Nursing for Longer Dynamic Video Synthesis (ICLR 2025 · similarity 0.90)
- Mind the Time: Temporally-Controlled Multi-Event Video Generation (CVPR 2025 · similarity 0.88)
- Before It Fades: Reinforcing Temporal Representations at Inference Time in VideoLLMs (NeurIPS 2026 · similarity 0.88)
- SlowFocus: Enhancing Fine-grained Temporal Understanding in Video LLM (NeurIPS 2024 · similarity 0.88)
- Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Search (NeurIPS 2025 · similarity 0.88)
- CineOrchestra: Unified Entity-Centric Conditioning for Cinematic Video Generation (NeurIPS 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
