Research map / Video / Video generation
V.I.P. : Iterative Online Preference Distillation for Efficient Video Diffusion Models
Jisoo Kim, Wooseok Seo, Junwan Kim, Seungho Park, Sooyeon Park, Youngjae Yu
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 SFT regularizer to diffusion DPO (ablation) | V.I.P. on AnimateDiff · VideoScore temporal consistency | score: 2.437 -> 2.513 | Improved |
| add SFT regularizer to diffusion DPO (ablation) | V.I.P. on VideoCrafter2 · VideoScore text alignment | score: 2.471 -> 2.518 | Improved |
| add SFT regularizer to diffusion DPO (ablation) | V.I.P. on AnimateDiff · VideoScore dynamic degree | score: 2.744 -> 2.695 | Hurt |
| use step-by-step online iterative distillation instead of offline (ablation) | V.I.P. on AnimateDiff · VideoScore dynamic degree | score: 2.670 -> 2.695 | Improved |
| use step-by-step online iterative distillation instead of offline (ablation) | V.I.P. on VideoCrafter2 · VideoScore text alignment | score: 2.483 -> 2.518 | Improved |
| use ReDPO instead of SFT distillation | V.I.P. on VideoCrafter2 · VideoScore text alignment | score: 2.505 -> 2.518 | Improved |
| use ReDPO instead of SFT distillation | V.I.P. on AnimateDiff · VBench semantic score | score: 74.7 -> 76.8 | Improved |
| use ReDPO instead of SFT distillation | V.I.P. on VideoCrafter2 · VBench semantic score | score: 72.2 -> 73.9 | Improved |
| apply ReDPO to stage 2 pruned model | VideoCrafter2 · VideoScore average | score: 2.608 -> 2.623 | Improved |
| apply ReDPO to stage 3 pruned model | AnimateDiff · VideoScore average | score: 2.566 -> 2.568 | Improved |
| raise SFT weight from 1e6 to 1e7 (ablation) | ReDPO on VideoCrafter2 stage 2 · VideoScore temporal consistency | score: 2.617 -> 2.403 | Hurt |
| lower SFT weight from 1e6 to 1e2 (ablation) | ReDPO on VideoCrafter2 stage 2 · VideoScore temporal consistency | score: 2.617 -> 2.292 | Hurt |
Closest papers by meaning
- T2V-Turbo-v2: Enhancing Video Model Post-Training through Data, Reward, and Conditional Guidance Design (ICLR 2025 · similarity 0.88)
- T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback (NeurIPS 2024 · similarity 0.88)
- VideoDPO: Omni-Preference Alignment for Video Diffusion Generation (CVPR 2025 · similarity 0.88)
- VideoGuide: Improving Video Diffusion Models without Training Through a Teacher's Guide (CVPR 2025 · similarity 0.88)
- DenseDPO: Fine-Grained Temporal Preference Optimization for Video Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Diffusion-DRF: Free, Rich, and Differentiable Reward for Video Diffusion Fine-Tuning (NeurIPS 2026 · 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.
