The model is trained on photorealistic images of fit and attractive females. It should work on all types of captions but the training dataset did not have any nudity.