Research map / Video / Video generation
VDE: Training-Free Accelerating Rectified Flow Model via Velocity Decomposition and Estimation
Junwen Tan, Jinglin Liang, Hongyuan Chen, Shuangping Huang
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 |
|---|---|---|---|
| use VDE-fast instead of EasyCache-fast | VDE · FLUX.1 [dev] 512x512 | LPIPS: 0.3197 -> 0.1997 | Improved |
| use VDE-slow instead of EasyCache-slow | VDE · FLUX.1 [dev] 512x512 | LPIPS: 0.2793 -> 0.1243 | Improved |
| use VDE-slow instead of EasyCache-slow | VDE · Qwen-Image 512x512 | LPIPS: 0.1445 -> 0.0691 | Improved |
| use VDE-fast instead of EasyCache-fast | VDE · Qwen-Image 512x512 | LPIPS: 0.4239 -> 0.1096 | Improved |
| use VDE-slow instead of EasyCache | VDE · Wan2.1-1.3B video | LPIPS: 0.0952 -> 0.0554 | Improved |
| use VDE-fast instead of EasyCache on VBench | VDE · Wan2.1-1.3B video | VBench (%): 80.49 -> 80.43 | Hurt |
| increase anchor interval from 1 to 5 (ablation) | VDE · FLUX.1 [dev] 512x512 | LPIPS: 0.0801 -> 0.2168 | Hurt |
| increase anchor interval from 1 to 5 (ablation) | VDE · FLUX.1 [dev] 512x512 | speedup: 1.69x -> 3.22x | Improved |
| replace extrapolated alpha, beta with true values (ablation) | VDE · FLUX.1 [dev] 512x512 | LPIPS: 0.1198 -> 0.0874 | Improved |
| replace reused orthogonal direction with true u_t (ablation) | VDE · FLUX.1 [dev] 512x512 | LPIPS: 0.1198 -> 0.0132 | Improved |
| increase resolution from 256x256 to 1024x1024 | VDE-slow · Qwen-Image | LPIPS: 0.1036 -> 0.0541 | Improved |
| increase resolution from 256x256 to 1024x1024 | VDE-slow · Qwen-Image | speedup: 2.07x -> 2.06x | No clear effect |
Closest papers by meaning
- Accelerating Rectified Flow Models via Trajectory-Aware Caching (NeurIPS 2026 · similarity 0.87)
- FlowTurbo: Towards Real-time Flow-Based Image Generation with Velocity Refiner (NeurIPS 2024 · similarity 0.87)
- FlowDiffuser: Advancing Optical Flow Estimation with Diffusion Models (CVPR 2024 · similarity 0.87)
- Improving Progressive Generation with Decomposable Flow Matching (NeurIPS 2025 · similarity 0.86)
- Improving the Training of Rectified Flows (NeurIPS 2024 · similarity 0.86)
- FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis (CVPR 2024 · similarity 0.86)
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.
