Research map / Video / Video generation
Towards One-step Causal Video Generation via Adversarial Self-Distillation
Yongqi Yang, Huayang Huang, Xu Peng, Xiaobin Hu, Donghao Luo, Jiangning Zhang, Chengjie Wang, Yu Wu
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 adversarial self-distillation (ASD) training (ablation) | few-step student · VBench 1-step | Total Score: 78.13 -> 80.65 | Improved |
| add adversarial self-distillation (ASD) training (ablation) | few-step student · VBench 1-step | Semantic Score: 69.31 -> 76.15 | Improved |
| add adversarial self-distillation (ASD) training (ablation) | few-step student · VBench 2-step | Total Score: 82.61 -> 83.28 | Improved |
| add first-frame enhancement (FFE) at inference (ablation) | few-step student · VBench 1-step | Total Score: 78.13 -> 83.04 | Improved |
| add first-frame enhancement (FFE) at inference (ablation) | few-step student · VBench 1-step | Semantic Score: 69.31 -> 79.95 | Improved |
| add first-frame enhancement (FFE) at inference (ablation) | few-step student · VBench 2-step | Total Score: 82.61 -> 83.80 | Improved |
| add FFE on top of ASD (ablation) | few-step student · VBench 1-step | Total Score: 80.65 -> 83.89 | Improved |
| add FFE on top of ASD (ablation) | few-step student · VBench 2-step | Total Score: 83.28 -> 84.32 | Improved |
| use ASD+FFE instead of retrained Self Forcing | Self Forcing · VBench 1-step | Total Score: 80.62 -> 83.89 | Improved |
| use ASD+FFE instead of Self Forcing at 4 steps | Self Forcing · VBench 4-step | Total Score: 84.31 -> 84.38 | No clear effect |
| set ASD loss weight alpha to zero (ablation) | ASD loss · VBench 1* and 2* step | Total Score | Improved |
| add ASD loss to DMD training (ablation) | DMD · DMD loss during training | DMD loss variance: 0.01963 -> 0.00258 | Improved |
Closest papers by meaning
- OSV: One Step is Enough for High-Quality Image to Video Generation (CVPR 2025 · similarity 0.91)
- SF-V: Single Forward Video Generation Model (NeurIPS 2024 · similarity 0.91)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.90)
- Diffusion Adversarial Post-Training for One-Step Video Generation (ICML 2025 · similarity 0.90)
- Equilibrium Forcing: Adaptive Video Generation Without Noise Conditioning (NeurIPS 2026 · similarity 0.89)
- Autoregressive Adversarial Post-Training for Real-Time Interactive Video Generation (NeurIPS 2025 · 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.
