Well, version 1.0 sure took a turn away from where I started. But this is a much better place to end up. I actually want to use this LoRA. Output is fun and adorable. The bokeh is pretty severe, but I'm enjoying that too. Maybe I'll try for an alternative version at some point that has longer depth of field. The model really embraces complexity - to the point of duplicating heads and little characters in odd places sometimes. A somewhat higher CFG (5 or so) seems to help most the time, though my typical spot is just under 4. You shouldn't need any keyword. But if you feel you do, say you want a clay sculpture illustration.

Files with SHA256: dea294e1b04ade579155c360d9bb4169dba43f4b305ea9b74cd2a53b4cf8896b