I retrained Nautilus from Azur Lane as a LoCon model with a few new images added to her dataset and made her flexible this time, however it kind of messes certain backgrounds like beach ones and her 'Afternoon Stretches' (trigger tag 'bunny') costume isn't well replicated since the folder for it had very few images compared to her default one, and in general she's more obscure than Mimori Mizuha from Blue Archive therefore her entire dataset had fewer images than her (don't get me started on the next two I'll retrain from the same game sharing my birthdate...).

Nautilus (Azur Lane) LoCon

I wanted to retrain this character from Azur Lane using Prodigy optimizer and some few new images added to her dataset. Seems results are a bit improved over DAdaptation I trained with the first time months ago. However said dataset is really small, so don't expect a lot of variety and some bleeding from dataset images to pop up frequently unless you don't use these tags in negative prompt if you generate her with the default costume: leg up, feet up, cane, holding cane, hand up, arm up, pointing, pointing up

Files with SHA256: 6b5d6a650f060cd231d391c022ce2ee86f3c0594b41e59adc524f3e282d543de