To jest bardzo ciekawe w kontekście ‘MMaDA-Parallel: Multimodal Large Diffusion Language Models for Thinking-Aware Editing and Generation’