MUNI is a framework for any-to-any generation that jointly trains modality-specific encoders, expressive decoders, and a single shared flow-based prior. It enables coherent generation between image, text, and audio modalities by using a routed training objective to align shared latents.