Paris 2.0 is the first video generation model pre-trained through decentralized computation, building on Paris 1.0 (the first open-weight Decentralized Diffusion Model). It demonstrates that temporally coherent video generation is achievable without a monolithic GPU cluster. In low-resolution text-to-video training, it cuts Fréchet Video Distance (FVD) from 561 to 279 (~2x improvement) compared to a monolithic model trained on matched compute.