LayerT2V: A Unified Multi-Layer Video Generation Framework

LayerT2V introduces a breakthrough multi-layer video generation framework that produces editable layered video components (background, foreground layers with alpha mattes) in a single inference pass. The system addresses professional workflow limitations of current text-to-video models by enabling semantic consistency across layers and introduces VidLayer, the first large-scale dataset for multi-layer video generation.

• →LayerT2V generates multiple semantically consistent video layers (background and foreground with alpha mattes) in one inference pass, unlike existing methods that only output final composited videos.
• →The framework uses temporal dimension serialization to jointly model multiple layer representations on a shared generation trajectory.
• →VidLayer dataset represents the first large-scale dataset specifically designed for multi-layer video generation training and evaluation.
• →The system employs a three-stage training process: alpha mask VAE adaptation, joint multi-layer learning, and multi-foreground extension.
• →Extensive experiments show LayerT2V significantly outperforms existing methods in visual fidelity, temporal consistency, and cross-layer coherence.