Paris 2.0: A Decentralized Diffusion Model for Video Generation
2026-05-25 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionMachine Learning
AI summaryⓘ
The authors introduce Paris 2.0, a new video generation model trained using decentralized computing, meaning many computers work together instead of one big supercomputer. It builds on their earlier Paris 1.0 model, which showed that images could be generated this way, but making smooth videos was still a challenge. Paris 2.0 improves the quality of generated videos significantly compared to a traditional single-model approach while using the same total computing power. Their results show better video quality scores and closer matching of video content to text prompts.
decentralized computationvideo generationdiffusion modeltext-to-videoFrechet Video Distance (FVD)CLIP text-video similarityaesthetic scoreGPU clusterParis 1.0temporal coherence
Authors
Ali Rouzbayani, Bidhan Roy, Marcos Villagra, Zhiying Jiang
Abstract
We present Paris 2.0, the first video generation model pre-trained through decentralized computation. Its training recipe builds upon Paris 1.0 (arXiv:2510.03434), the first ever open-weight Decentralized Diffusion Model (DDM), which showed that image generation can be trained without a monolithic GPU cluster. However, temporally coherent video generation had remained an open problem under decentralized training, and Paris 2.0 closes it. In low-resolution text-to-video training, against a monolithic model trained on the same data under a matched total compute budget, Paris 2.0 cuts Frechet Video Distance (FVD) from 561.04 to 279.01, a ~2.0x improvement, and lifts CLIP text-video similarity and aesthetic score.