Omni-Diffusion Unifies Multimodal Understanding and Generation | KnowAI Space