I made my Surface of the Moon / Lunar Surface / 月面 LoRA and found it to unstable, only working some of the time and giving wildly wrong results most of the time. I can't figure out why it doesn't like to work. Since the training data was exclusively public domain images, it's up for DL on the model page. I'm requesting people who know more about LoRA training than me to give me some feedback and help me figure out why it's like this and how I can improve it in the future (short of waiting till humanity goes back to the moon and gets more photos!).