Mannequin faceless egghead Reviews

doesnt work at all, on low weights most of my checkpoints still draw faces, on high weights i get probably almost the original training images of mannequins on pedestal with highheels plus backgrounds and sometimes mutated people, even when not using the mannequin trigger since of course since its a real word most checkpoints use their own data on top then. next time create clean images without backgrounds from all sides with different poses without clothes, then maybe generate some backgrouds for some of em and mix those into the datases, properly caption em and use a non real world trigger word, then train on standardized checkpoint instead of your seemingle pretty special one to make it work on most checkpoints, and when done test on other checkpoints than the one it was trained on, results will bemuch better. and for god sake dont just name a lora "lora".... and since you made a new version, you should make it on 1.5 then too to really see and compare the improvement. yours is on xl which by itself understands the concept much better even without lora so the results would be flawed and you cant see your actual improvements as good.