SteerVTE: Seamless Video Text Editing with Style and Glyph Control

SteerVTE is a novel framework for video text editing that utilizes a frozen video diffusion model to achieve precise modifications while maintaining stylistic consistency. It incorporates a lightweight text context adapter with a style encoder and dual-granularity glyph encoders, along with a glyph-aware spatial-focal loss and a three-stage training curriculum. This approach, supported by the SteerVTE-1M dataset of one million triplets, significantly enhances text accuracy, style consistency, and temporal coherence compared to existing baselines, making it a valuable tool for practitioners in video editing and AI-driven content creation.