This is an experimental style lora intended to produce a semi-realistic digital concept art style. It is trained entirely on artificial data from a Z-Image style that I found pleasing. It isn't quite where I want it, but it's come close on Anima. It's not super stable, but with some prompt adjustment, you can get around model biases that tend to fight against more realistic styles.Trigger word: p4interlyHelpful words: (realistic:0.5)Adding 'realistic' at medium weight can help pull it away from anime style when you're using danbooru tags that pull it towards anime for some reason. You can just go normal weight on that if you want, but it can take things too far. I also find it useful to avoid 1girl and 1boy and lean into the natural language style that Anima's language interpreter can handle.

Files with SHA256: 2c75bfb9b652c80ab909c7e27d8512485a9379b94e83efc0b57f8058eb2b544c