Show-o2: Improved Native Unified Multimodal Models | Read Paper on Bytez