T-Rex2 is a paradigm-shifting object detection technology that can recognize a wide range of objects, from everyday to esoteric, without task-specific fine-tuning or massive training datasets. It combines vision and text prompts, giving it powerful zero-shot capabilities, and can be widely applied to various scenarios of object detection tasks. T-Rex2 integrates four components: image encoder, visual prompt encoder, text prompt encoder, and box decoder. It follows the end-to-end design principles of DETR and covers various application scenarios. T-Rex2 achieved the best performance on four academic benchmark tests: COCO, LVIS, ODinW, and Roboflow100.