Research map / Video / Video generation
Video-SVD: Efficient Video Diffusion via Orthogonal Basis Composition
Zhang Wan, Yu Li, Tianze Huang, Haochen Li, Juan Cao, Sheng Tang
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 RoPE positional residual to basis-only attention (ablation) | Video-SVD · Wan2.1-1.3B | VBench: 71.53 -> 76.51 | Improved |
| add content residual to basis-only attention (ablation) | Video-SVD · Wan2.1-1.3B | VBench: 71.53 -> 78.24 | Improved |
| add both residual streams (ablation) | Video-SVD · Wan2.1-1.3B | VBench: 71.53 -> 81.76 | Improved |
| add both residual streams (ablation) | Video-SVD · Wan2.1-1.3B | speedup: 1.95x -> 1.75x | Hurt |
| use uniform instead of adaptive Top-5% sampling for Stage 2 (ablation) | Video-SVD · Wan2.1-1.3B | latency (s): 283.6 -> 268.0 | Improved |
| use adaptive Top-5% instead of uniform sampling (ablation) | Video-SVD · Wan2.1-1.3B | PSNR: 27.50 -> 27.53 | No clear effect |
| use uniform instead of random sampling (ablation) | Video-SVD · Wan2.1-1.3B | PSNR: 27.46 -> 27.50 | No clear effect |
| use checkpoint-adaptive bases instead of Linformer | Video-SVD · Wan2.1-1.3B | PSNR: 20.8 -> 27.5 | Improved |
| use Video-SVD instead of SVG on high-motion videos | SVG · pre-softmax reconstruction error | MSE: 0.250 -> 0.090 | Improved |
| increase basis number K (ablation) | Video-SVD · Wan2.1-1.3B | quality vs speed | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VORTA — VORTA: Efficient Video Diffusion via Routing Sparse Attention (NeurIPS 2025) · better on 3 of 3 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
- Sparse Video-Gen: Accelerating Video Diffusion Transformers with Spatial-Temporal Sparsity (ICML 2025 · similarity 0.91)
- Faster Video Diffusion with Trainable Sparse Attention (NeurIPS 2025 · similarity 0.90)
- VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion (NeurIPS 2026 · similarity 0.88)
- Bidirectional Sparse Attention for Faster Video Diffusion Training (NeurIPS 2026 · similarity 0.88)
- Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation (NeurIPS 2025 · similarity 0.88)
- Efficient Video Diffusion Models via Content-Frame Motion-Latent Decomposition (ICLR 2024 · 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.
