Language Model Backbone and Super-Resolution

Image, video, and audio are tokenized into a shared space, enabling a decoder-only model to generate outputs and control tasks via input-output token patterns.