Adaptive Region-Based Image Editing via Dynamic Bounding Box Optimization for Text-Guided Local Manipulation

In recent years, text-guided local image editing techniques have gained increasing popularity. This paper presents a novel text-guided adaptive region-based image editing method that eliminates the need for external masks, depth maps, or other prior conditions. Building upon existing pre-trained text-to-image models, our approach introduces adaptive bounding boxes as intermediate variables to automatically identify optimal editing regions. The study demonstrates that the proposed method not only achieves strong feasibility but also maintains compatibility with image generation model, enabling editing under complex textual prompts. Through user studies comparing our approach with state-of-the-art techniques, experimental results confirm its superior editing performance.