Trouble getting my Lora to work? Start here.
If you love my work: Consider supporting the cause ;)
Another Lora, that likes to break all your bones, if you are not careful with it.
Tested well on: CyberrealismXL, Cheyenne, PonyXL
mediocre result on: JuggernautXL, ZavychromaXL
Usage Tips:
Clean up your prompt, loras. Rebuild using a few of the tags below.
Set your Steps to 50 - 100% higher, than you normally would (This model fights against everything what sdxl and most checkpoint think a person should look like anatomically, you sometimes can see whole limbs changing position after step 30)
Try to get to a good set of prompts that seem to go in the right direction and then reroll till you arrive at a point where anatomy is mostly correct.
Do upscale, but be careful with the denoise. If set to high, some kneecaps may reorient and arms dislocate
Useful Taglist:
(these were used in training, some work well in prompts, some don't, some conflict with each other)
general pose encouragement: bent over, arms up | arms behind back, legs together | legs apart
perspective: from front, from angle, from side, from behind, from above, from below
looking directions: up, down, to the side, at viewer
bondage related: (color) + rope, bound + (bodypart), chained + (bodypart), (bodypart) + cuffs, armbinder, spreader bar, chains
Good Luck, Folks!
Description
FAQ
Comments (13)
Base model is SDXL!
Not again! i always forget this setting. thanks for commenting.
Plays well with Jib Mix, fyi.
Love your loras in general.
Any chance you could make some for Pony XL?
i've had a look at it. every lora i've tested worked ok with pony as normal sdxl versions. i don't see the need really.
@Joschek Could always be my settings, but I've had trouble getting the loras for Yoke, Panelgag, latex_skirt, pluggag, spidergag and better leashes to work properly with pony. Either way, appreciate your work :)
@shamalamalam maybe i'll give it another go, when i find the time. Do YOU know how training settings translate from sdxl to pony? are you supposed to train with clip skip 2?
@Joschek No idea I'm afraid
What about vae?,also I'm getting anime results every time even if I put anime or cartoon on negative and also you didn't put the cfg scale
this is a lora. vae and cfg will mostly depend on your checkpoint. if you are getting anime results, that will also be because of the checkpoint you are using. this is trained on a realistic dataset.
@Joschek with your Lora's I use cyberealistic xl and I asked about the cfg scale you used on your image example with the same checkpoint
@kd231002573 5 cfg is a good starting point. if you are getting anime results with cyberrealistic and just this lora, then something weird is going on.-
@Joschek I know,is pretty annoying,it's happening with like 40 steps or around,and not only with it,also happening with sdxxxl lol, anyway I will try in 5 and waiting for you to make more Loras ;)