The flux.2 klein algorithm has poor aesthetics, so this model was trained to improve it.

flux.2 klein 的美学效果较差，因此训练这个lora来改进它。

It's still trained using a multimodal approach, and the editing results are excellent, capable of performing many editing functions. Once the bug in the online training of the tensor's edit model is fixed, I will continue to improve this LoRa model.

依旧是多模态方式训练，编辑效果优秀，能完成很多种编辑功能。等tensor的edit模型在线训练bug修复，我还会继续改进这个lora的。

Regular: 0.4-1， 20-30 steps, cfg4-6, Euler + simple，Recommend: 0.6-1