
测试了主流的几个XL大模型,泛化性还可以,古今中外各种类型基本也都可以跑出来,
采样器我用的dpm++2m karras,步数25就可以跑出不错的效果,权重根据大模型及绘画主体不同可自由调节,可以先从1逐渐累加到1.5,
不过参数没有那么死板,还请大家自由发挥~
I tested several mainstream XL models, and the generalization was okay.All kinds of paintings are worth trying.
I used the dpm++2m karras sampler, and 25 steps could produce good results. The weights could be adjusted freely according to the different models and drawing subjects. You can start from 1 and gradually increase to 1.5.
But the parameters are not so rigid, please feel free to play around~
描述:
训练词语: fen9zik4i
名称: fen9zik4i.safetensors
大小 (KB): 223103
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success