Type | |
Stats | 231 0 |
Reviews | (33) |
Published | Sep 30, 2023 |
Base Model | |
Training | Steps: 1,000 Epochs: 20 |
Usage Tips | Clip Skip: 2 |
Trigger Words | Yuzu Yukimoto hairband animal on head |
Hash | AutoV2 ACA3692F2E |
Hello! First LoRa I ever make, and probably the last!
I take requests on Pixiv to generate images of certain characters, and one of them was Yuzu Yukimoto, from the anime Sunohara-Caretaker. Which had no model! So I made one using holostrawberry's simple guide for making basic models!
How good is it? I only had 7 proper images to work with! ...it's ass.
Not the best, but at least it's better than nothing!
During my testing, a weight of 0.7 was enough to make decent images. Base activation tags are "Yuzu Yukimoto, hairband, animal on head". The main tag, 'Yuzu Yukimoto' is generally consistent on giving her the basic features, like body type, height, hair and face, but will still sometimes change her hairstyle or body type. Again, only 7 images to work with!
The 'Animal on head' tag is completely random! Only a few of the images I got gave her her funny little bird-thing on her head, while most only had the hairband, so I apologize for that! Don't worry! Inpainting with a simple prompt like "Yellow bird on head" worked for me before!
For her clothing. Most clothing tags should work fine! I didn't do extensive testing on many outfits, but the ones I did didn't have trouble at all.
Finally, I also didn't test too much of LoRa flexibility! I made some images with a few other models... don't ask which ones. Definitely don't check my profile either, if you understand!