Research map / Video / Video generation
EFlow: Fast Few-Step Video Generator Training from Scratch via Efficient Solution Flow
Dogyun Park, Yanyu Li, Sergey Tulyakov, Anil Kag
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 Mean-Velocity Additivity regularizer (ablation) | EFlow PDG-SFM · Kinetics-700 | FVD: 185.2 -> 146.7 | Improved |
| replace Sana attention with GLGA (ablation) | FM VideoDiT · Kinetics-700 | FVD: 217.7 -> 191.3 | Improved |
| replace softmax with GLGA attention (ablation) | FM VideoDiT · Kinetics-700 | FVD: 188.9 -> 191.3 | No clear effect |
| add 75% token dropping to Sana attention model (ablation) | Sana VideoDiT · Kinetics-700 | FVD: 217.7 -> 456.1 | Hurt |
| add 75% token dropping to GLGA model (ablation) | GLGA VideoDiT · Kinetics-700 | FVD: 191.3 -> 189.7 | No clear effect |
| add sliding window attention to linear attention (ablation) | GLGA (300K iters) · Kinetics-700 | FVD: 420.8 -> 352.9 | Improved |
| add input-aware gating to summed linear+window attention (ablation) | GLGA (300K iters) · Kinetics-700 | FVD: 352.9 -> 325.7 | Improved |
| add linear attention to sliding window attention (ablation) | GLGA (300K iters) · Kinetics-700 | FVD: 390.2 -> 352.9 | Improved |
| use PDG-SFM instead of standard SFM at 4 NFE (ablation) | EFlow (300K iters) · Kinetics-700 | FVD: 347.1 -> 322.2 | Improved |
| use PDG-SFM at 4 NFE instead of FM at 4 NFE (ablation) | EFlow (300K iters) · Kinetics-700 | FVD: 513.8 -> 322.2 | Improved |
| use PDG-SFM 4 NFE instead of FM 40 NFE (GLGA, token drop) (ablation) | EFlow · Kinetics-700 | FVD: 189.7 -> 185.2 | Improved |
| add path-drop guidance to cut training cost of CFG | EFlow · 480p video training throughput | speedup: 1.76x and 2.5x higher training throughput | Improved |
Closest papers by meaning
- Training-Free Efficient Video Generation via Dynamic Token Carving (NeurIPS 2025 · similarity 0.88)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.87)
- Optical-Flow Guided Prompt Optimization for Coherent Video Generation (CVPR 2025 · similarity 0.87)
- Faster Video Diffusion with Trainable Sparse Attention (NeurIPS 2025 · similarity 0.87)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.86)
- Bidirectional Sparse Attention for Faster Video Diffusion Training (NeurIPS 2026 · 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.
