Updated: Feb 12, 2024
character
Trained with 16 images (Almost screenshots) and the base model is AOM3.
Trigger word is "furukawa nagisa". You may want to emphasize some features by using the prompt, because the output can suffer from lack of them.
Works fine with 2D anime style models. Not very compatible with 2.5D or realistic style models. Maybe I can finally find a way to make a 2.5D version of it.
v1.1 is based on GhostMix V2.0(https://civitai.com/models/36520/ghostmix).
V1.1 is not necessarily better than V1.0, but may benefit from a larger training set.
V2.0: Same training data but with the new optimizer Prodigy and base model. I recommend this version because it should have better generalization and robustness.
Hi I'm greenhand at making LoRA. This is my first LoRA so please know that it's not very good enough. However I love and respect CLANNAD and Nagisa very much. Hope you like this anime and this LoRA, too!