This is the base model I use for almost all of my generation. It is a merged checkpoint combining killedmyself's Useless Animerge and the kemiao version of newlifefexfztty761's Cute Yuki Mix.
Description
- Initial release
FAQ
Comments (12)
Is there a reason for it not being safetensor?
Not certain. The original file is a safetensor, so I'll have to see at what point during the upload it changed.
I hope you will reupload it. The model looks interesting and I want to try it.
for what reason its not pruned?
I just decided to leave the model as is since it worked fine and I had the space on my PC and I haven't pruned a model before.
any instructions if using hi-res fix with this?
No, unfortunately. I've tried using a couple different models with this for hi-res fix, but they always seem to heavily effect the style no matter how late I apply them.
@reggiemccorthy1457 so no hi-res is recommended? and also what about denoising? any recommended instructions for that?
@pokezilla39966 I'd recommend no hi-res. If I may be honest, I've never tried denoising on any image generation. I'm afraid I can't offer you any advice on that.
@reggiemccorthy1457 righto, you should give it a shot sometime, though, the result may surprise you :D.
For me personally, best option is to use hi-res steps at 10, and denoising set at 0.3. Use whatever model but in my personal opinion, use the hi-res model Remacri: https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri. (Make sure to put the downloaded file in stable diffusions ESRAGEN file, this is really important), then select it in the hi-res model select page on stable (you may have to refresh it for it to appear), Then bump the resolution to what you want. It's simple really, after all is set, generate away.
Also if you want, using ADetailer isn't much different, just download then activate it, set 1st area onto: face_yolov8n.pt (ignore hyperlink), then move onto the second one, not adjusting anything else on the first, only making sure the settings of control net guidance start is at 0 and guidance end is at 1, and control net strength is a 1, use these for the second area aswell.
for the second one, use mediapipe_face_mesh_eyes_only. for the prompt area, use whatever color eyes the character has, for example: (green eyes is on the character, type in prompt area, Emerald green eyes).
Hopefully this helps you out :D
@pokezilla39966 This is great, thanks! I've considered trying before, but I could never find a decent description of how to go about it and could never get and good results. I'll definitely have to try this out. I appreciate the tips.
@reggiemccorthy1457 no problem, let me know if you have any questions or issues.
The denoising essentially clears up static around the faces, and anything that looks unnatural. If using your older images and running them through denoising, i will be curious as to how good they look with hi-res fix and denoising compared to how you originally generated them. :D (No offense by the way, your older images are still good, just saying they might look better once run through denoising and hi-res fix) :D