The aim of this model is to increase the understanding of longer and more complex prompts.

This model is highly experimental.

Recommended weight of ~0.75

Training Notes

The model is trained with very long and complex image captions. The intent was to see if by training with very detailed descriptions, I could increase the prompt understanding.

Sometimes I feel like it's really increasing the prompt understanding, but it may certainly be bias. A lot more testing and training is needed!

The dataset is now available, as well as a brief article on the creating it:

https://civarchive.com/articles/14046/the-caption-process-of-daiversity-detailer