Research map / Video / Video generation
VPO: Aligning Text-to-Video Generation Models with Prompt Optimization
Jiale Cheng, Ruiliang Lyu, Xiaotao Gu, Xiao Liu, Jiazheng Xu, Yida Lu, Jiayan Teng, Zhuoyi Yang, Yuxiao Dong, Jie Tang, Hongning Wang, Minlie Huang
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 preference optimization (DPO) after SFT (ablation) | VPO-SFT · MonetBench, CogVideoX-2B | Overall: 3.59 -> 3.76 | Improved |
| add preference optimization (DPO) after SFT (ablation) | VPO-SFT · MonetBench, CogVideoX-5B | Overall: 4.01 -> 4.15 | Improved |
| add text-level feedback to video-level feedback (ablation) | VPO w/o TL FDBK · MonetBench, CogVideoX-2B | Overall: 3.72 -> 3.76 | Improved |
| add text-level feedback to video-level feedback (ablation) | VPO w/o TL FDBK · MonetBench, CogVideoX-5B | Overall: 4.12 -> 4.15 | Improved |
| add text-level feedback to video-level feedback (ablation) | VPO w/o TL FDBK · Query alignment, CogVideoX-2B | Aligned (%): 93.0 -> 94.6 | Improved |
| add text-level feedback to video-level feedback (ablation) | VPO w/o TL FDBK · Query alignment, CogVideoX-5B | Aligned (%): 92.8 -> 94.8 | Improved |
| add video-level-only preference optimization after SFT (ablation) | VPO-SFT · Query alignment safety (unsafe rate, CogVideoX-2B) | Unsafe (%): 0.8 -> 2.0 | Hurt |
| use VPO instead of GPT-4o few-shot prompt rewriting | GPT-4o Few-Shot · Query alignment | Aligned (%): 86.4 -> 94.6 | Improved |
| use Promptist image prompt optimizer instead of original query | Original Query · MonetBench, CogVideoX-2B | Overall: 3.27 -> 2.87 | Hurt |
| use VPO instead of GPT-4o few-shot rewriting | GPT-4o Few-Shot · MonetBench, CogVideoX-5B | Overall: 4.03 -> 4.15 | Improved |
| apply prompt optimizer trained on CogVideoX-2B to another video model | GPT-4o Few-Shot · MonetBench, Open-Sora 1.2 | Overall: 3.07 -> 3.18 | Improved |
| use VPO instead of VBench Long Prompts | VBench Long Prompts (2B) · VBench Multiple Objects | score (%): 63.81 -> 70.17 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Prompt-A-Video — Prompt-A-Video: Prompt Your Video Diffusion Model via Preference-Aligned LLM (ICCV 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- The Devil is in the Prompts: Retrieval-Augmented Prompt Optimization for Text-to-Video Generation (CVPR 2025 · similarity 0.91)
- SafeSora: Towards Safety Alignment of Text2Video Generation via a Human Preference Dataset (NeurIPS 2024 · similarity 0.89)
- Neuro-Symbolic Evaluation of Text-to-Video Models using Formal Verification (CVPR 2025 · similarity 0.89)
- PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought (CVPR 2024 · similarity 0.89)
- EvalCrafter: Benchmarking and Evaluating Large Video Generation Models (CVPR 2024 · similarity 0.89)
- Boosting Text-to-Video Generative Model with MLLMs Feedback (NeurIPS 2024 · 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.
