Type | |
Stats | 745 |
Reviews | (52) |
Published | Jun 12, 2023 |
Base Model | |
Training | Steps: 15,000 |
Hash | AutoV2 7A5826B6B6 |
"Dr. Beverly Crusher is two spheroid regions approximately 20 centimeters in diameter." -Enterprise D Computer
Version 2 Notes:
This is significantly better than version 1. Less faces in dataset but more action poses and works in every checkpoint I've tried. The likeness is a lot stronger and you don't need to include much of anything in your prompts. In case you run into problems, feel free to throw in "blue uniform" or "red hair" to strongly adhere, but as before, there's no trigger words.
Version 1 Notes:
This LoRA was an absolute nightmare to make and I'm not sure why. This is the best one out of several variations of datasets and different numbers of iterations. Any training improvement comments are welcome, the more technical the better.
Anyways, this LoRA does occassionaly switch genders, so throwing in "woman" in the prompt can resolve that. All training data was on the blue starfleet uniform with a few images with the additional overcoat. It heavily favors without the overcoat, but you may get traces. Try using: "blue starfleet uniform" in the prompt to heavily enforce the former.
No trigger word needed.
Any model except RealisticVision. For some reason this checkpoint will give some small, but noticable blue/green discoloration. You are welcome to use it, just be advised I'm aware of this issue.