WanPE: Towards Cinematic Prompt Enhancement for Modern Text-to-Video Generation
WanPE is a 397B-parameter prompt enhancer built to turn user prompts into shot-level cinematic plans for video models.
The paper says it was trained on 1.05M real-world videos and uses video-grounded reverse construction plus SC-GRPO to preserve user intent over time. Its benchmark, WanPEval, covers 5- to 30-second clips with about 11K blind pairwise human assessments. When used with Wan3.0, WanPE beat raw prompts by 10.66-18.84 preference points on 5-15 second videos and by 50.86 points at 30 seconds. HF Daily Papers' note
The paper says it was trained on 1.05M real-world videos and uses video-grounded reverse construction plus SC-GRPO to preserve user intent over time. Its benchmark, WanPEval, covers 5- to 30-second clips with about 11K blind pairwise human assessments. When used with Wan3.0, WanPE beat raw prompts by 10.66-18.84 preference points on 5-15 second videos and by 50.86 points at 30 seconds. HF Daily Papers' note
score 6