WanPE: Towards Cinematic Prompt Enhancement for Modern Text-to-Video Generation

2026-09-24

AI Summary

This study addresses the challenge of precisely planning long-duration, multi-shot cinematic sequences in text-to-video generation by proposing a director-level narrative model with 397 billion parameters. Methodologically, this work introduces video reverse engineering to automatically generate shot plans and proposes a semantic-consistent GRPO reinforcement learning algorithm. By integrating large language models with multimodal alignment techniques, the approach ensures instruction fidelity across shots and over extended temporal dimensions. Experimental results demonstrate that the proposed model significantly outperforms commercial counterparts in generating 5- to 15-second videos, while achieving substantial improvements in human preference scores for 30-second long-form videos, thereby establishing a new benchmark in the text-to-video generation domain.

Abstract

Video generation begins in text space by authoring a cinematic screenplay, then materializes into pixels. As contemporary video generators scale to 30 seconds and faithfully follow complex conditions, the textual prompt largely directs the production, planning how actions, camera trajectories, lighting, and sound unfold across multi-shot sequences. In this paper, we present WanPE, a 397B-parameter prompt enhancement model trained on 1.05M real-world videos to master director-level cinematic planning. WanPE formulates shot-level cinematic plans via video-grounded reverse construction and employs Semantic-Consistency GRPO (SC-GRPO) to faithfully preserve user requirements across shots and over time. To benchmark this capability, we curate WanPEval, a human-annotated testbed covering durations from 5 to 30 seconds across varying intent granularities, supported by approximately 11K blind pairwise assessments. When powering Wan3.0's video generator, WanPE-397B boosts human preference over raw user prompts by 10.66-18.84 points at 5-15 seconds and by a dramatic 50.86 points in the 30-second arena. Ablation studies show that reverse construction demonstrates clear superiority over forward rewriting, while SC-GRPO robustly preserves semantic fidelity across model scales. Ultimately, WanPE leads all evaluated commercial offerings at 5-15 seconds and remains competitive with Seedance 2.5 at 30 seconds.

Problem

• text-to-video generation
• prompt enhancement
• cinematic planning
• multi-shot sequences
• semantic consistency

Innovation

• Prompt Enhancement
• Reverse Construction
• SC-GRPO
• Text-to-Video Generation
• Cinematic Planning