Research map / Video / Video generation
Swap Attention in Spatiotemporal Diffusions for Text-to-Video Generation
Wenjing Wang, Huan Yang, Zixi Tuo, Huiguo He, Junchen Zhu, Jianlong Fu, Jiaying Liu
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 |
|---|---|---|---|
| add swapped spatiotemporal cross-attention (3D window) (ablation) | LDM-based video diffusion · WebVid-10M val (1K samples) | FVD: 566.16 -> 475.09 | Improved |
| add swapped spatiotemporal cross-attention (3D window) (ablation) | LDM-based video diffusion · WebVid-10M val (1K samples) | CLIPSIM: 0.3070 -> 0.3107 | Improved |
| swap Q and K,V roles between spatial and temporal features instead of fixing them (ablation) | Swap-CA (3D window) · WebVid-10M val (1K samples) | FVD: 490.60 -> 475.09 | Improved |
| use 3D window attention instead of global attention (ablation) | Swap-CA · WebVid-10M val (1K samples) | FVD: 485.86 -> 475.09 | Improved |
| use Swap-CA instead of 3D spatiotemporal WSA (ablation) | LDM-based video diffusion · WebVid-10M val (1K samples) | FVD: 500.49 -> 475.09 | Improved |
| use Swap-CA instead of Tune-A-Video attention (ablation) | LDM-based video diffusion · WebVid-10M val (1K samples) | FVD: 717.34 -> 475.09 | Improved |
| use window size 8x3x6 instead of 8x1x3 (ablation) | Swap-CA · WebVid-10M val (1K samples) | FVD: 525.91 -> 475.09 | Improved |
| increase temporal window from 8 to 16 (ablation) | Swap-CA · WebVid-10M val (1K samples) | FVD: 475.09 -> 487.08 | Hurt |
| add HD-VG-130M to training (ablation) | Swap-CA model · WebVid-10M val (1K samples) | FVD: 475.09 -> 429.75 | Improved |
| fine-tune on HD-VG-40M higher-quality subset (ablation) | Swap-CA model · WebVid-10M val (1K samples) | FVD: 429.75 -> 418.40 | Improved |
| train on HD-VILA-100M instead of HD-VG-130M (ablation) | Swap-CA model · WebVid-10M val (1K samples) | FVD: 429.75 -> 692.99 | Hurt |
| caption with mPLUG-2 instead of BLIP-2 (ablation) | Swap-CA model · WebVid-10M val (1K samples) | CLIPSIM: 0.3046 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- LaVie — LaVie: High-Quality Video Generation with Cascaded Latent Diffusion Models (IJCV 2024) · better on 2 of 2 shared results
- CogVideo — CogVideo: Large-scale Pretraining for Text-to-Video Generation via Transformers (ICLR 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Vivid-ZOO: Multi-View Video Generation with Diffusion Model (NeurIPS 2024 · similarity 0.89)
- EIDT-V: Exploiting Intersections in Diffusion Trajectories for Model-Agnostic, Zero-Shot, Training-Free Text-to-Video Generation (CVPR 2025 · similarity 0.89)
- Grid Diffusion Models for Text-to-Video Generation (CVPR 2024 · similarity 0.89)
- VideoGrain: Modulating Space-Time Attention for Multi-Grained Video Editing (ICLR 2025 · similarity 0.89)
- OpenVid-1M: A Large-Scale High-Quality Dataset for Text-to-video Generation (ICLR 2025 · similarity 0.89)
- A Recipe for Scaling up Text-to-Video Generation with Text-free Videos (CVPR 2024 · similarity 0.89)
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.
