sd-model-trained-lora-Os

Hi,guys! 这是一个以石窟中的菩萨为母题制作的Lora，还在实验中，大家在使用后可否给我提点意见，感激不尽！ 此Lora搭配CounterfeitV30大模型效果还行，如果想生成菩萨的样子，触发词可用“Bodhisattva”。 我在测试中发现，采样器使用“DPM++ SDE Karras”最佳，迭代步数需要开到30，CFG最好是10。 关于这个Lora的权重，实测中0.2-0.4风格较为多变和自由，0.6-1则出现菩萨的样貌最为明显。当然，在搭配其他Lora使用后，效果也还不错。

(以下是机器翻译内容：) This is a Lora made with the bodhisattva in the grotto as the mother theme, still experimenting, can you give me some advice after using it, thank you very much! This Lora works fine with CounterfeitV30 large model, if you want to generate the appearance of Bodhisattva, the trigger word can be used "Bodhisattva". I found in the test, the sampler using "DPM++ SDE Karras" best, the number of iterative steps need to open to 30, CFG preferably 10. Regarding the weight of this Lora, the actual measurement of 0.2-0.4 style is more variable and free, and 0.6-1 appears the most obvious Bodhisattva-like. Of course, after using with other Lora, the effect is not bad.