Even today some progress was released on parallelizing WAN video generation over multiple GPUs. LLMs are way easier to split up.