This version is bringing the model closer to gryphon anatomy, which is the main goal of this lora. The trigger words (s4grif, gryphon) should be enough to get it to spit out 4 legged gryphons with a much cleaner look. It still is not enough to completely override illustrious's baked in human bias, but it made a pretty big dent, and can sometimes even make things that look kind of nice.
It also works with other models down stream from illustrious, so give that a try if so inclined.
Changes: I cleaned up my messy tags while keeping the same number of images. I then did training batch of 10 per lora check point. I would go between setting the learn rate high and then setting it low, all while tracking how broken the lora was starting to look. I stopped at a place that felt good. I also increased the size of the lora to see what kind of effect it would have.
Description
This version is bringing the model closer to gryphon anatomy, which is the main goal of this lora. The trigger words (s4grif, gryphon) should be enough to get it to spit out 4 legged gryphons with a much cleaner look. It still is not enough to completely override illustrious's baked in human bias, but it made a pretty big dent, and can sometimes even make things that look kind of nice.
It also works with other models down stream from illustrious, so give that a try if so inclined.
Changes: I cleaned up my messy tags while keeping the same number of images. I then did training batch of 10 per lora check point. I would go between setting the learn rate high and then setting it low, all while tracking how broken the lora was starting to look. I stopped at a place that felt good. I also increased the size of the lora to see what kind of effect it would have.