Research map / Video / Video generation
DOLLAR: Few-Step Video Generation via Distillation and Latent Reward Optimization
Zihan Ding, Chi Jin, Difan Liu, Haitian Zheng, Krishna Kumar Singh, Qiang Zhang, Yan Kang, Zhe Lin, Yuchen Liu
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 consistency distillation to VSD (ablation) | DOLLAR student · VBench long prompts | Total Score: 80.08 -> 80.65 | Improved |
| add latent reward fine-tuning to VSD+CD (ablation) | DOLLAR student · VBench long prompts | Total Score: 80.65 -> 82.57 | Improved |
| use 5 teacher denoising steps in CD instead of 1 (ablation) | VSD+CD student · VBench long prompts | Total Score: 78.92 -> 80.65 | Improved |
| use homogeneous v-prediction instead of heterogeneous x-prediction (ablation) | VSD+CD student · VBench long prompts | Total Score: 80.05 -> 80.65 | Improved |
| add consistency distillation to VSD (ablation) | DOLLAR student · Vendi Inception diversity | Vendi score: 1.91 -> 2.14 | Improved |
| distill with VSD only versus teacher (ablation) | VSD student · Vendi Inception diversity | Vendi score: 2.34 -> 1.91 | Hurt |
| use LRM instead of DDPO for reward tuning (ablation) | VSD student · VBench long prompts, HPSv2 reward | Total Score: 81.29 -> 81.96 | Improved |
| use LRM instead of DDPO for reward tuning (ablation) | VSD student · VBench long prompts, PickScore reward | Total Score: 81.84 -> 81.71 | No clear effect |
| use long prompts instead of short prompts (ablation) | VSD4+LRM · VBench | Total Score: 78.12 -> 81.71 | Improved |
| use long prompts instead of short prompts (ablation) | VSD1 · VBench | Total Score: 80.70 -> 80.62 | No clear effect |
| vary student sampling steps 1, 2, 4 under VSD (ablation) | VSD student · VBench long prompts | Total Score: 80.62 / 80.94 / 80.08 | Mixed |
| distill 50-step teacher into 4-step DOLLAR student | DOLLAR · VBench long prompts | Total Score: 80.25 -> 82.57 | Improved |
Closest papers by meaning
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.90)
- T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback (NeurIPS 2024 · similarity 0.90)
- OSV: One Step is Enough for High-Quality Image to Video Generation (CVPR 2025 · similarity 0.90)
- Simple and Fast Distillation of Diffusion Models (NeurIPS 2024 · similarity 0.88)
- InstructVideo: Instructing Video Diffusion Models with Human Feedback (CVPR 2024 · similarity 0.88)
- Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation (NeurIPS 2025 · similarity 0.87)
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.
