tip1：当步数为28时，lora权重最好在0.5~0.6之间

TIP2：当lora权重在0.55时，步数最好在20~60之间

如果和别的Lora混合，例如与“鹅蛋脸lora“，两个lora权重都要下降0.1~0.15，并且步数也要下降到20左右。

本人电脑性能有限，大家可自行测试。

lora训练基础参数

大模型：chilloutmix

vae:vae-ft-mse-840000-ema-pruned

采样方法：DDIM

Tip1: When the step count is 28, the optimal Lora weight is between 0.5 and 0.6.

Tip 2: When the Lora weight is 0.55, the optimal step count is between 20 and 60.

If mixing with other Loras, such as "ovoid face face Lora," both Lora weights should be reduced by 0.1-0.15, and the step count should also be reduced to around 20.

Due to limited computer performance, please conduct tests on your own.

Lora training basic parameters:

Large model: chilloutmix

VAE: vae-ft-mse-840000-ema-pruned

Sampling method: DDIM