This paper introduces a novel two-stage object detection framework aimed at addressing the challenges of fine-grained object detection. The proposed framework unifies the strengths of Grounding DINO and CLIP to achieve nuanced and precise fine-grained object detection. A key innovation is the introduction of a class-subclass dictionary, which serves as a critical component for the seamless integration of new classes into the detection framework. This dictionary facilitates the framework’s adaptability to evolving class landscapes, providing a scalable and efficient solution for dynamic object detection tasks. Unlike conventional approaches that often require extensive model retraining to accommodate new class sets, this method eliminates such limitations by adopting a more flexible and adaptive strategy. Grounding DINO is utilized to generate robust and accurate region proposals, showcasing its remarkable generalization abilities across a wide range of object classes. These region proposals are then refined using CLIP, which assigns precise fine-grained labels to the detected objects. The class-subclass dictionary acts as a bridge between these two components, facilitating seamless communication and integration.To evaluate the framework’s performance, prompt sensitivity experiments were conducted using two benchmark datasets: NABirds and FGVC-Aircraft. The results highlight the effectiveness of the proposed method, with impressive mAP@0.5 scores of 47.44 on the NABirds dataset and 40.95 on the FGVC-Aircraft dataset. These findings underscore the superior performance of the framework in fine-grained object detection tasks, setting a benchmark for future research in this area.