>>3036027I have about 500 of these, use barely two dozen with some regularity. Guess my top is Nanameda Kei, but he barely does anything on merges.
imo the key thing is to run your tests on the base model or a very weak merge, minimize quality tags and use a simple prompt. Makes it easier to pick out their differences if they aren't all biased towards the same thing. Just look at Raita.
