Lora-scale: 0.5-0.7
Trigger Words: akagi miria, short hair, black hair, brown eyes, two side up
All my models are free, and I'm happy to accommodate any requests—just send me a message on Civitai chat. If you'd like to support my work, you can do so through Ko-fi!
Description
Improving quality and character
FAQ
Comments (5)
Seems a bit overcooked, what were your training settings?
According to my overall assessment and feedback from some users, version 2.0 does not generate images with accurate characters. In 2.1, I kept the settings the same as the old version, but I made some adjustments to the dataset to help it learn better.
Why do you think it overcooked? Do you mind explain it more clearly
@duongve13112002 the fact that it works best at so low weight, it seems you either have a very big dataset or it was over trained (several repeats compared to the dataset size)
@SuzuHoshi All characters in this series have been trained with a large dataset that includes all costumes, along with some noisy data to prevent the model from overfitting. Furthermore, I tested it with my method and found that the model works more stably compared to the old version about the style and character's appearance, so you don't need to worry too much.