LoRA of Kursk from Azur Lane

Recommended parameters:

• LoRA weight: 0.6-0.8

• Sampling method: Euler a or DPM++ 2M Karras or SDE

• Sampling steps: 20-40

• CFG Scale: 4-11

• Upscaler: depend on model that you use

  UPDATE
  I did V2 (2 versions with deferent training parameters) based on new dataset but i have no option to test it, so i will be grateful if someone can test it and send a results:
  LINK

描述:

Data set is 18 imagines. It's enough to make simple LoRA, but not enough for perfect result.
It will be improved in future!

训练词语: sn-kursk