Trained on a subset of the Flickr faces dataset.

1. Entirely new, hand written captions.

2. Zero NSFW Images in Dataset

3. Only photographs (Flicker faces has images of dolls those where pruned)

You can describe in natural language or Vit tagging.

1. An ugly mature female with a square face.

2. mature, square face, 1girl, ugly

This lora address some of the issues with zero shot image classification. Such as the 60-80% accuracy on race. The fact that it does not ascribe aesthetics or face characteristics. As well as other issues.