Dynin-Omni: Omnimodal Unified Large Diffusion Language Model

Jaeik Kim, Woojin Kim, Jihwan Hong +8

We present Dynin-Omni, the first masked-diffusion-based omnimodal foundation model that unifies text, image, and speech understanding and generation, together with video understand…