Some actual information about the model

I wanted to learn about full-scale fine tuning and see what would happen if I attempted to tune/improve a model within a rather small subset of prompt-space (I guess). I started with bigLoveXL (https://civarchive.com/models/897413/big-love) and trained it on a few thousand images of a particular style (think social media more than anything else). I ran it for a few days. I did not keep the log files, so I do not know the exact step count, but maybe 50k ish.

The differences between bigLoveXL and this model are small for many prompts, but for some you will see a big difference. I am also happy with the small differences you see when you do not see major differences. I will most likely try this again.

I used joycaption, so you might get some mileage out of prompting like joycaption captions.

All in all, bigLoveXL is just as good/better and this was always meant to be a experiment for myself. Some people asked me to post the model, so here it is.

Due to lack of time when posting the model, all the examples are done without any detailing or fixing. Just two phases.