>>47313883>>47313831It made zero difference. If you want to grab it and see for yourself before I delete it
https://rentry.org/vtairecipes#new-super-gekij%C5%8Dban-shinseiki-dai-rant%C5%8D-hll4p3-you-can-not-proompt-viii-epoch-hd-remix-irys-turbo095ada70b8https://huggingface.co/grugger/chubas/blob/main/models/hll4-p3-ep8/ULTRA_SGSDRHLL4p3YC(N)PVEHDR%26IRyS_XL.safetensorsIt's the same exact thing, but mixed step by step without any HLL and then AD1.0 of the whole thing at the end
7th_anime_v2_C-1995-0760-3559-FP32 - fixed the CLIP encoder, nothing else
7th_v2_A_v2_C_WS_25 - weighted sum merge, A=7th_v2_A, B=7th_v2_C, 0.25
Counterfeit-V2.5_[7th_v2_A_v2_C_WS_25]_WS60, A=Counterfeit_V2.5, B=Previous Step, 0.6
expmixLine_v20_[Counterfeit-V2.5_[7th_v2_A_v2_C_WS_25]_WS60]_WS60, A=expmixLine_v20,B=Previous Step, 0.6
ULTRA_SGSDRHLL4p3YC(N)PVEHDR&IRyS_XL == [emL_v20_[Counterfeit-V2.5_[7th_v2_A_v2_C_WS_25]_WS60]_WS60]_hll4-p3-ep8_AD1
So no, HLL doesn't "stack" when you weighted sum a bunch of merges that had HLL AD1.0'd into them.
