>>48542548I think what you've done is very reasonable merge. Out of interesting, I'm going to try a multi-step merge that starts with the HLL model itself instead of an existing mix. Of course, the most autistic option would be Bayesian merger, but I know nothing about it. I know there's an anon who messed around with it.
>>48542243And just for fun, I used my architecture LoRAs and that prompt and seed. Naturally, some of the rrats can no longer be described as having "chubby abs".
https://files.catbox.moe/wamhon.7z 