The training dataset consists of 50+ anime characters with daily humorous styles (you may recognize them at a glance).
As I was struggling to find an avatar for myself, I unexpectedly found that it matched well with Lora, a fictional character.
No specific keywords are required (you can try "close up"), and setting the weight to around 0.8 will produce significant results.
The recommended size is 544x544 (a square is fine, as it is an avatar).
日本动漫风格头像生成
训练图集是50+日常搞笑风的动漫(估计眼熟的一眼认出来了)
由于我苦于没有头像炼的一个lora
发现和人物lora意外的匹配?!
不需要特别的关键词()可以试试close up),权重拉到0.8左右就会有明显的效果
推荐尺寸:544X544(方形就行,因为是头像嘛)
最后推荐一个可以在线使用这个模型的地方:https://tusi.art/models/605765713598188624