
This version includes 2 LoRA, one for the character+outfits and another for Ayakura Juu's style: https://twitter.com/haino
CivitAI does not allow multiple LoRA in one version so I've put the style LoRA in the "training data" zip file. I've included example images both with and without the style; refer to the prompts to see how to use the LoRa together. All images are generated using Grapefruitv4.1.
The character LoRA was trained on NovelAI with 2000 steps and the style LoRA with 3000 steps. Best strength for both LoRA is about 0.5-0.7. Style does not require trigger words. I find the generations are most accurate when including "sash" and "pouch" tags in the prompt. Also try increasing the strength of the trigger words e.g. (holo:1.5).
Feel free to share your images.
描述:
训练词语: holo,holoBlueDress,holoBrownDress,holoGreyShirt,holoPinkShirt,holoCoat
名称: ayakurajuu.zip
大小 (KB): 33581
类型: Training Data
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
名称: holo.safetensors
大小 (KB): 36982
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success