During April 2025, Zhang Tianyu developed the LiteRAWFormer model architecture for the Shubhamsaboo/OmniGen2 repository, focusing on raw image processing. He designed a lightweight, transformer-based foundation by implementing core components such as LayerNorm, Attention, and FeedForward blocks, along with utilities for tensor manipulation and channel shuffling. Using Python and PyTorch, Zhang established a modular architecture that supports efficient processing of raw images and lays the groundwork for future features like raw image super-resolution. His work demonstrated depth in model architecture and image processing, providing a robust base for advanced user-facing enhancements in the OmniGen2 project.