switched to SDXL instead of illustrious as well as added a few more refs in varied positions to see if it helps produce more varied outputs
edit: much like v1, this one works best at 0.5 to 0.7 if making normal sized characters
added a few more refs for it to train on. still slowly building on this.