For a long time, the detection of diabetic retinopathy has always been a great challenge. People want to find a fast and effective computer-aided treatment to diagnose the disease. In recent years, the rapid development of the deep learning makes it gradually become an effective technique for the analysis of medical images. In this paper, we propose a method to deal with diabetic retinopathy images with generative caption technique of images to generate a simple sequence to explain the abnormal contents in fundus images. The generative technique of images is a generative model based on a deep recurrent architecture that combines convolution neural network (CNN) which is currently state-of-the-art for object recognition and detection with long-short-term-memory (LSTM) which is applied with great success to machine translation and sequence generation, and that can be used to generate natural sentences describing an image. The target of the model in training is to maximize the likelihood of the target description sentence given from the training images. The model built on dataset DIARETDB0, DIARETDB1 and Messidor can achieve good performance and generate fluent sequences. In addition, the experimental results show that the accuracy of diagnosis for individual abnormal discoveries is up to 88.53% and the diagnosis accuracy is more than 90%.