我制作的角色Lora侧重于泛用性,不会将角色原本的服装固定学习,只保留角色样貌的还原,所以角色原本的服装需要提示词才能触发,而且不会特别还原,但是它可以很好得更换衣服以及涩涩。
如果你需要还原服装建议使用其他大佬制作的Lora。
这个Lora不需要提示词就能工作,如果需要生成帽子建议加上hat,提示词。
V2版本优化了拟合度,修复了一些错误学习的问题。
当其他tag过多,或者其他Lora权重高时可能会导致形象没还原,在不修改权重的情况下可以加上这些tag来稳定生成:twintails, drill hair, twin drills, blonde hair,
The Lora I made will not focus on restoring the character's costume, but on making it easier for the character to change and remove the costume.
This Lora does not need a hint word to work if it does not generate a hat suggestion plus a hat, hint word.
This Lora does not need a hint word to work, if you need to generate hat suggestion plus a hat hint word
The V2 version optimizes the fit and fixes some mislearning issues.
When there are too many other tags, or other Lora weights are too high, Lora may not take effect. You can add these tags to stabilize the generation without modifying the weights: twintails, drill hair, twin drills, blonde hair,
推荐权重/weight: 0.8~1
例图模型/Sample graph model:cetusMix-v3.5
采样方法(Sampler):DPM++SDE Karras
放大算法(Hires.fix):R-ESRGAN 4x+Anime6B (0.3)