I just removed two very/to prominent images from the original dataset and tried a few programs to create the LoRA for FLUX models.

Actually I'm only testing on forge UI but what I've read, if it works in forge, it should normally also work on comfy.

I would say that the resulting LoRA does play pretty well with other LoRAs and most checkpoints.

Experiment with the weight of the LoRA, so try values from :0.4> to about :1.4>

Above 1.4 almost all checkpoints always gave me black images.

The only working (as intended) Trigger is "pointy breasts", but it may also already give you desired results without this extra trigger.

For the next version, a new dataset of much more and even better images and descriptions/prompt-keywords is in progress, but I think it will take some time to get it done until I'm satisfied.
And to be honest, it's not really fun to investigate 4k video for 4k video skip frame for frame for the best shot of a pair of boobs especially only for such a dataset.

But I think FLUX needs it just to enable us all to create realistic images of for example simply beach scenes that please our eyes.