Type | |
Stats | 209 4,028 382 |
Reviews | (35) |
Published | Mar 5, 2024 |
Base Model | |
Training | Steps: 2,500 Epochs: 10 |
Usage Tips | Strength: 1 |
Hash | AutoV2 CB09500478 |
SDXL Model of Squigly, from the fighting game Skullgirls. Similar to my Ms. Fortune LORA I had a 1.5 LORA but didn't see any for SDXL. Trained for about 2500 steps on roughly 200 images with 5 repeats. Images were taken from official in-game sprites as well as fan art. While the dataset is almost entirely SFW, the LORA can produce NSFW images as well. Has a propensity to generate vibrant backgrounds that I can't quite explain. You can prompt "simple background" in the positives or "blue background" in the negatives to get rid of this. Other than that, Squigly's partner Leviathan seems to appear in generations almost randomly. As there is no tag to prompt that character, consider this a limitation of this LORA, and possibly AI in general.
I recommend using around 1.0 strength, but if you're finding you have trouble combining it with other LORAs, lower the strength to 0.8 or 0.7. If you'd like the dataset used to train this LORA, please reach out to me via a comment or direct message on this platform.