During April 2026, contributed to the bytedance-iaas/sglang repository by integrating the ERNIE-Image model for text-to-image generation. This work involved designing the model architecture, implementing encoders, and configuring inference pipelines to support diffusion-based image synthesis. Enhanced the input processing pipeline with prompt enhancement features, enabling higher quality outputs and more flexible workflows for experimentation and customer-facing demos. Leveraged expertise in Python, PyTorch, and deep learning to establish a production-ready foundation for configurable inference. The integration addressed the need for scalable, modular text-to-image generation, laying the groundwork for future improvements in computer vision and natural language processing applications.