Mochi 1

An advanced, open-weight video generation model developed by Genmo, designed to bridge the gap between abstract text prompts and high-fidelity, physically consistent video outputs. Operating on an open Asymmetric Diffusion Transformer (AsymmDiT) architecture with roughly 10 billion parameters, it is explicitly trained to master real-world physics, fluid dynamics, and complex multi-object interactions. Mochi 1 generates crisp, high-resolution clips characterized by smooth temporal consistency and minimal motion blurring. By treating video data as fluid multi-dimensional tokens, it ensures that characters and background environments remain spatially coherent even during extreme camera movements. This makes Mochi 1 a premier tool for animators, digital artists, and software developers seeking a highly reliable, localized video synthesis engine.