The style of Kamiya Yuu, the author of No Game No Life. His defining feature is the use of vibrant colors and gradients, which I'm not sure how well this LoRA captures, but oh well. If anyone were to train a better alternative, I'd be delighted.
The dataset was the set of Gelbooru posts tagged with kamiya_yuu. Version A uses (slightly preprocessed) gelbooru tags is now unpublished by CivitAI, version B uses CivitAI autotagging instead. The training parameters also differ a bit, I feel like I still haven't got a good feel for what different values result in. I think A is noticeably stronger because it used alpha=4, dim=16 vs B with alpha=16, dim=16. I'll appreciate any advice on how to do better in the future. For now, I decided to publish both versions together. The trigger words are optional in both cases: in A they were present in images, but weren't pinned with "keep tokens", while in B they weren't present at all, but the model's prior knowledge might still have a positive influence on the style.
Also, I tagged it as ???? but I think the style should do fairly well for SFW works too, considering those form the majority of the dataset.
If you like the style, please support the original author
Retrained on a smaller, hand-picked dataset with other changes in training method inspired by my success with the Spidu style LoRA. This version still doesn't quite live up to my expectations, but it feels better than the previous one.
1. 轉载模型僅供學習與交流分享，其版權及最終解释權归原作者。
2. 模型原作者如需認領模型，請通過官方渠道联系海藝AI工作人員進行認證。我們致力於保護每一位創作者的權益。 點擊去認領
