Based on anything4.5, tested with anything4.5, cetusMix and pastelMix.更二次元的lora模型。

ASOUL贝拉Bella-LORA

You need to use ribbon, hat to distinguish clothes.由于训练集里混合了多套服饰，所以需要用ribbon, hat等trigger来区分。The advantage of not limiting the costume is that it has stronger generalizability, and the disadvantage is that the reproduction of the default costume and pico is not as good as the pico model.Better support for Hires. fix.不限制服饰，优点是有更强的泛用性，缺点是对于默认服饰和pico的还原度不如pico模型。对于Hires. fix的支持更好。

Files with SHA256: c166915aa2f93112a760d9db1259de230295e93ebe99a72612526f721af7ef07