This checkpoint was built using thousands of my own photos and merged with a bunch of other forgotten base datasets over time to keep it stable. What came out wasn’t planned, but it ended up being one of the most accurate models I’ve used for character LoRAs.

I kept trying to make it more flexible or creative (not that this one is incapable, i've made some wild shit), but every time I did, it lost some of that accuracy. So before I messed with it too much, I wanted to share this version as-is. That said, hands are hit or miss.

It works really well with concept LoRAs too, so you can still get creative without breaking character identity LoRAs. Everything in the showcase is made without any LoRA's at all, but I'll be posting many in the weeks to come with dozens of LoRAs I've been training.

If you arent using LoRAs and keep your descriptions kinda ambiguous ('long hair' instead of "long brown hair') you will get varied appearances very reliably as my datasets contain many facial variations.