share the details using catbox or other service that doesn't strip metadata please 🤙
Discussion
Positive: (starry, space)
masterpiece, highres, absurdres, newest, 1girl, solo, wolf ears, brown wolf tail, fringe trim, (pointy hair:1.2, spike hair:1.5, sidelocks), long hair, brown hair, green eyes, eyelashes, looking at viewer,
Negative: bad quality, worst quality, lowres, jpeg artifacts, bad anatomy, bad hands, multiple views, signature, watermark, sketch, flat color, ugly, fat, plump, muscular, abs, blurry eyes, wrinkled skin, fat, straight hair, nsfw, tanlines, skinny, anorexia, child, greyscale, extra ears, straight hair, ponytail, ahoge, blurry eyes,blank eyes, multicolored hair, (frame:1.6), outframe, outline, framed, window, door, corridor, blonde, (camera frame, picture frame, out of frame, framed)
Sampling: Kohaku_LoNyu_Yog
Schedule: Karras
Upscaler: RealERSGAN_x4plus_anime_6B
🤔 spo in already illustrious trained with that... could be sampling and schedule and holy negatives LOL do you also have enabled edsn 3133769? :'p if yes doesn't do much so put in 0, enable discard penultimate sigma. And avoid skip early cfg greater than 0 it helps on certain conditions.. A lot stuff you gonna need to test 🤔
LOL believe or not, I slowly added each negative, so they get me the results I'm looking for, but I need to do that with positive and see if fix the issue next time
It stopped now for some reason, I just changed the model, it is catTowerNoobaiXL_v11, changed to V14, maybe it was the model
check then if is not based on vpred all vpred is very experimental and doesn't play well with skip early and you need to enable zero snr terminal for vpred only. the advanced sampling extension is not needed now.