An image description method includes obtaining an image; obtaining an image description model and a description prompt text, the description prompt text indicating an image description rule for performing image description by the image description model; inputting the description prompt text and the image into the image description model; and performing image description on the image based on the description prompt text through the image description model, to obtain a first image description text that conforms to the image description rule. The description prompt text is obtained by correcting an initial prompt text corresponding to the image description rule based on a first text difference between a sample image description text and a predicted image description text. The sample image description text is obtained based on a sample image and the image description rule. The predicted image description text is obtained by the image description model performing image description on the sample image based on the initial prompt text. The initial prompt text is initialized based on the image description rule.
Full Text
What is claimed is: