This LoRA is trained using game assets from the F&C game "White Breath ~ with faint hope~". The character artist is Takashi Hashimoto who later on worked on "Yosuka No Sora" hence similar art style. (his prior works "Pia Carrot e yokoso 3" and subsequence works have different art style).
The version 1 was trained using the original VGA resolution images. It tends to give a blurry images.
Version 2 was trained with larger dataset with upscaled images. It seems to give the sharper output comaparing to version 1.
Description
Initial version
FAQ
Comments (2)
To improve the blurry output try upscaling the source image to 1024. If you are using A1111 you can pick scale to -> width -> 1024. Make sure the upscaled image looks ok. If it doesn't, try a different upscaling method or exclude the image. I would also try removing the trigger word in your training.
Thank you for your suggestions. I'll be working on it.