This is a sequal to my previous model MikuFactoryXL, in my opinion it is better in everyway. Much less blurry, better line art, better hands, more creativity, better details etc. Is also has a different art style however, which I prefer more. It was created by merging MikuFactoryXL and AltMikuFactoryXL which I haven't uploaded to CivitAI yet but can be found at https://huggingface.co/nonetrix/MikuFactoryXL and https://huggingface.co/nonetrix/AltMikuFactoryXL and finally https://huggingface.co/nonetrix/MikuFactoryXL-2
Description
Improved art style (subjective)
Improved hands
Improved creativity (subjective)
Improved background details
FAQ
Comments (11)
Looks amazing, thank you for sharing
any chance FP16?
How do I quantize it exactly? I mislabeled it honestly lol. Not sure why ComfyUI saved it as fp32 either
@iamjustheretodow6140 sorry I don't know. I only used a1111 for it and only few times :/ (Now I use Comfy and I never do it)
@iamjustheretodow6140 "One important thing you should note is that models are merged and saved in the precision that is used for inference on your hardware so usually it will be 16 bit float. If you want do do merges in 32 bit float launch ComfyUI with: –force-fp32"
Maybe this can help.
@DrPunk Oh I ran in CPU mode and made 200GB Swap file, using all the VRAM on my AMD GPU makes it crash prone. Makes sense why it ended up like that now. Likely better way to merge models ComfyUI is too dumb to unload models
Sorry I said figured out how to do it but seems model doesn't work after doing it my bad lol...
Will upload fp16 pruned version soon but struggling to get it to work, if anyone knows how to make it work please tell me lol. I only found https://github.com/lopho/stable-diffusion-prune which seems to output a non working model :/
Sorry for uploading a fp32 model in first place, this model was created on CPU since my AMD GPU shits itself randomly and reboots my whole PC at random usually if I use too much VRAM... Thanks Lisa Su
a1111 has special category for any model merging/changing. I don't know how it looks now but i remember I was able to change model into "pruned". It was easy but be careful with options. I used it in SD 1.5 era and have 0 idea how it work now but it was easy thanks to clean UI.
@DrPunk I miss automatic1111 but runs poorly on my AMD GPU for some reason, should just buy used 3090 at this point lol
SD.Next has convert model tab, i use that (6700xt user here)