Research map / Video / Video generation
Exploring Data-Free LoRA Transferability for Video Diffusion Models
Yuchen Wang, Wenliang Zhong, Lichen Bai, zikai zhou, Shitong Shao, Bojun Cheng, Shuo Chen, Shuo Yang, Zeke Xie
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 restoration to arbitration (full CASA vs arbitration only) (ablation) | CASA · Steamboat-Willie on FastWan2.1-T2V-1.3B | CSD (%): 77.68 -> 81.49 | Improved |
| add arbitration to restoration (full CASA vs restoration only) (ablation) | CASA · Steamboat-Willie on FastWan2.1-T2V-1.3B | Quality Score: 1.23 -> 1.58 | Improved |
| add arbitration to restoration (full CASA vs restoration only) (ablation) | CASA · Jinx-v2 on FastWan2.1-T2V-1.3B | Quality Score: 1.45 -> 1.51 | Improved |
| add restoration to arbitration (full CASA vs arbitration only) (ablation) | CASA · Jinx-v2 on FastWan2.1-T2V-1.3B | CSD (%): 67.56 -> 70.28 | Improved |
| apply CASA instead of direct LoRA reuse | LoRA transfer to distilled Wan2.1 1.3B models · Steamboat-Willie on FastWan2.1-T2V-1.3B | Quality Score: 1.27 -> 1.58 | Improved |
| apply CASA instead of direct reuse on 14B targets | LoRA transfer to Wan2.1 14B distilled models · Steamboat-Willie-14B on FastWan2.1-T2V-14B | CSD (%): 63.47 -> 62.56 | Hurt |
| apply CASA instead of direct reuse on Jinx-v2 / Rolling Forcing | LoRA transfer · Jinx-v2 on Rolling Forcing | Quality Score: 2.69 -> 2.67 | No clear effect |
| use CASA instead of ProLoRA for data-free transfer | LoRA transfer · Steamboat-Willie on FastWan2.1-T2V-1.3B | CSD (%): 60.15 -> 81.49 | Improved |
| use CASA instead of ProLoRA | LoRA transfer · Jinx-v2 on FastWan2.1-T2V-1.3B | Quality Score: 1.63 -> 1.51 | Hurt |
| apply CASA on a different model family (HunyuanVideo-1.5) | CASA · Retro-Anime on HunyuanVideo-1.5-480P-T2V-CFG-Distill | CSD (%): 75.83 -> 77.52 | Improved |
| vary quantile thresholds qdom and qact (ablation) | CASA · Sensitivity analysis (Figure 8) | Quality / CSD | No clear effect |
Closest papers by meaning
- One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution (NeurIPS 2025 · similarity 0.86)
- Spectral Motion Alignment for Video Motion Transfer Using Diffusion Models (AAAI 2025 · similarity 0.85)
- CoDMD: Copula-aware Distribution Matching Distillation for Fast Video Generation (NeurIPS 2026 · similarity 0.85)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.85)
- Spectral Progressive Diffusion for Efficient Image and Video Generation (NeurIPS 2026 · similarity 0.85)
- LoRAtorio: An intrinsic approach to LoRA Skill Composition (NeurIPS 2026 · similarity 0.85)
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.
