Lora for tsukushi. Trigger word is tsukushi.
I recommend ADetailer or hi res fix if your faces are coming out bad at low res. Most of my sample images are using ADetailer.
v2
Different training set, different training parameters. A little different stylistically but should look more like her. Trained against perfect world. Use "glasses" to get her glasses.
v1
Original. Trained against fantexi v0.8
Description
FAQ
Comments (2)
Please make new hair style.
problem: all the (frameless, round circle, half-rim, nerdy, fashionable, etc...) eyeglasses will morph into her default glasses. this will make some outfit unsuitable for her glasses style, like the ball dress.
work around: use comfyUI "conditioning (concat)" node to separate the "glasses". use image2image, but the 2nd image don't use tsukushi lora. the glasses concat text prompt box, yellow node (clip), can be routed as usual to both image, but purple node (model) only the first image NOT the 2nd one, hopefuly image2image will somehow recognize the glasses shape,
proper solution (my pure speculation): future lora use "tsugla" to triger her default glasses. it didn't contain "glasses" so it shouldn't trigger anything else, only "her" glasses, not comon eye glasses or anything else. if you use "tsukushi_glasses" then I will be locked out from using "rimless eyeglasses"
I have no idea how to make lora, I could be wrong 100% cuz im only using not making lora. anybody smarter than me is welcome to comment my mistake below