Seaweed-7B: Cost-Effective Training of Video Generation Foundation Model

Challenging the bigger-is-better paradigm in video AI

Foundation models serve as the cornerstone of modern AI, trained on vast amounts of data with massive parameter counts. These models demonstrate strong generalization capabilities across various domains — from language processing to image understanding. But in video generation, the computational demands have become extreme, with models like MovieGen requiring 6,000+ NVIDIA H100 GPUs.