just for test, and proof i do this test - it merge 50/50 of pony and illustrious lora trained on same datasets, for look how good result be and how compatible it be, my result is:
1) quality is medium -> useless to do it;
2) i sure it kill flexibility too (even if i don't test it with multiple loras and long prompting i see problems even on small ones)

why i put it here: for people can look that you can get by this stupid merge idea without made 2 loras. (!it same dataset as on big one!)