Developed a performance-oriented feature for the kvcache-ai/sglang repository, focusing on optimizing the image encoding path within a machine learning workflow. The solution intelligently bypassed negative prompt encoding when the guidance scale was less than or equal to 1.0 or when no negative prompt was provided, thereby reducing unnecessary computation and improving throughput. This approach maintained output quality while enhancing resource efficiency, supporting scalable inference across diverse prompt and guidance settings. The work was implemented using Python and leveraged expertise in image processing and machine learning, with an emphasis on code quality and measurable improvements in processing speed and resource utilization.