SwitchCraft: Training-Free Multi-Event Video Generation with Attention Controls

Published in CVPR 2026, 2026

SwitchCraft is a training-free framework for multi-event video generation. It introduces Event-Aligned Query Steering (EAQS) to steer frame-level attention to align with relevant event prompts, and Auto-Balance Strength Solver (ABSS) to adaptively balance steering strength for temporal consistency and visual fidelity. Experiments demonstrate substantial improvements in prompt alignment, event clarity, and scene consistency over existing baselines.

Recommended citation: Q Xu, C Song, Y Cai, C Zhang. (2026). "SwitchCraft: Training-Free Multi-Event Video Generation with Attention Controls." CVPR 2026.
Download Paper