note that our tools may result in inaccurate result, such as human and flowers, animals may be recognized as containing texts, can you provide the prompt which are used to instruct blip-2 to regenerate caption considering our result.