There is a new version available: https://civarchive.com/models/315602?modelVersionId=354037

A test project on a pornographic dataset.

This time "deepthroating dildo", "Fingering Pussy" and "drinking pee" have been added, "fucking machine". ("" are the main trigger words, I've tried to pick the trigger words that don't overlap with other lora)

After testing, when the prompt is too complex, it is more likely to respond to the parts that are not related to the training content. So it is recommended to add trigger words directly after the roles in natural language. if using comfyUI, you may need to experiment more with the weight of the prompt.

The goal is to make the characters in the concept (looks, body type, ethnicity) easily controllable without being limited by the dataset (small number of images used).

Manually tagging most of the concepts that are already in the base model.

The example images were generated by two different models, basemodel used for training and juggernautXL, juggernautXLV9 is highly recommended and helps in correct human structure.

If there is a subsequent update, it will be a release that adds more concepts to this base after known issues have been resolved. It's up to me to get lora to understand the various parts of the male and female anatomy using reg images, maybe someone can teach me? Otherwise the quality of the images generated is too dependent on the choice of checkpoints.