WanPE: 397B Prompt Enhancer Boosts 30s Video Preference by 50.86 Points
Yubo Zhu · hf · 2026-09-25
WanPE is a 397B-parameter prompt enhancement model trained on 1.05M real-world videos for director-level cinematic planning in text-to-video generation.
- Method: Shot-level cinematic plans via video-grounded reverse construction, with Semantic-Consistency GRPO (SC-GRPO) preserving user requirements across shots and time
- Evaluation: Accompanied by WanPEval, a human-annotated testbed covering 5–30s durations, supported by 11K blind pairwise assessments
- Results: Powering Wan3.0's generator, human preference rises 10.66–18.84 points at 5–15s and a dramatic 50.86 points at 30s; leads all evaluated commercial offerings at 5–15s and stays competitive with Seedance 2.5 at 30s
- Ablations: Reverse construction clearly beats forward rewriting; SC-GRPO robustly preserves semantic fidelity across scales
More from Multimodal
- Sophon-Realtime-v1, a full-duplex speech-to-speech model, trends on Hugging Face — adventists-ai · 2026-11-11
- Asked Fable 5.2 to 'go wild' making a game, got a Pong clone — and loved it — ariG23498 · 2026-11-11
- Western town built with Fable 5.2: ray-traced weathering ages every board — ravisparikh · 2026-11-11
- Diffusion Week 2026 releases all 10 lectures on video generation, world models and more — LucaAmb · 2026-11-11
- Running Wan 2.2 on a RTX 3080 12GB: 5B Breaks Down, 14B Works Surprisingly Well — fixxxer2606 · 2026-11-11
- Indie dev teases Z-Image-based model 'on par with latest closed-source models' — darlens13 · 2026-11-11