Trained on 800 images generated on Midjourney v3. Unlike the original version of this training, I removed less artistically pleasing images every few epochs. I think this has given me an overall better result, however this isn't a general style that applies. This lora will effect your prompt based on what its generating and how similar it is to the content in the training data. As such, it can be all over the place. You can bend it closer to the actual Midjourney3 style by adding that term to the prompt for by increasing the strength over 1. In my limited testing I've found that around 1.5 strength the training data will start to overpower the prompt.

Still, just like the original version, I'll probably use it in conjunction with other lora to push the overall style more painterly, surreal, and artistic.