wxli318/PixelWizard

PixelWizard: Towards Efficient High-Fidelity Video Generation at Ultra-Large Spatial Resolution

PixelWizard is a high-resolution video generation model designed for efficient 2K/4K video synthesis. It decouples global structure modeling from high-resolution detail generation through spatial-temporal anchor modeling and anchor-guided synthesis. The model further uses Noise-Span Aligned Shortcut Training for fast few-step inference.

Citation

Paper: https://arxiv.org/abs/2605.25801