step-distillation

Salt++ 是一个两阶段后训练框架，通过 CausalSelf-Flow 和上下文对齐的自回归 DMD，解决少步流式音视频生成中教师强制与因果分布蒸馏的上下文错配问题，在 480p 4-step 设定下相对 OmniForcing 提升视觉质量 57%、运动质量 45%。

Flash-WAM introduces a modality-aware distillation method for world-action models, achieving real-time inference by compressing diffusion to a single step per modality, resulting in 23x speedup.

LongCat-Video-Avatar 1.5 is an upgraded open-source framework for audio-driven human video generation with improved lip synchronization, production-ready stability, and efficient 8-step inference.