Instruction-based image editing improves the controllability and flexibility of image manipulation using natural commands without elaborate descriptions or regional masks. However, human instructions are sometimes too brief for current methods to capture and follow. Multimodal Large Language Models (MLLM) show promising capabilities in cross-modal understanding and generating visual responses through LM. We investigate how MLLMs facilitate editing instructions and introduce MLLM-Guided Image Editing (MGIE). MGIE learns to derive expressive instructions and provides explicit guidance. The editing model jointly captures this visual imagination and performs manipulation through end-to-end training. We evaluate various aspects of Photoshop styling, global photo optimization, and local editing. Extensive experimental results demonstrate that expressive instructions are crucial for instruction-based image editing, and our MGIE can lead to notable improvement in automatic metrics and human evaluation while maintaining competitive inference efficiency.
