Worked on enhancing multimodal processing capabilities within the zjunlp/EasyEdit repository, focusing on improving how models handle both image and text inputs. The primary contribution involved updating the MEND and WISE methods to support a broader range of models, including Blip2, MiniGPT-4, Qwen2-VL, and LLavaOV. This work leveraged deep learning and machine learning techniques, utilizing Python and transformer architectures to enable more robust cross-model input handling. The enhancement aimed to provide richer user experiences and expand the system’s applicability to diverse multimodal tasks, reflecting a targeted and technically deep approach to integrating advanced processing across multiple model backends.