ERNIE: New data set; shouldn't have the halo/clipping issues the ZIT version has. Z IMAGE TURBO: This is a decent looking LoRA but I tried something that didn't really work very well. And it's only my 3rd zimage LoRA so I'm still learning how these things operate differently than SDXL models do. What I did was I created a bunch of SDXL images and then took them into photoshop, removing the backgrounds completely and saving the files as png before training the ZIT Lora. The result is that she has a bit of a halo effect around her, even when you place her into a setting. She feels a bit removed. It's a unique look but it wasn't intentional. I may revisit this lora but ZIT cost significantly more than the SDXL ones and I'm still learning how to write captions that work better than tags did.

Files with SHA256: 1b034f7bfa28e7dfcf74a67250ed3244961bb1574cd5379792599e3db1136c34