You could upload image in the visual variant of the model and ask to describe it. It’s different from Dall-E 3