A Lora for creating truly pale characters.
VERSION 1.0 NOW OUT: Training steps now 7500 for much higher versatility/reliability.
Recommended lora weight: 0.2 - 0.8
A considerable portion of the training dataset was comprised of alt models, so this lora does especially well at creating gothic/alternative and or tattooed characters! The main goal of the lora is to be able to generate very pale characters without having to use prompts like 'goth', 'gothic' as this tends to change the overall style of the image as well. In version 1.0, using gothic prompts can now create truly white/grey skin!
The lora was trained using only realistic images for creating realistic/photorealistic characters. From some preliminary testing, the lora does work somewhat with more artistic models and will just add its own realism elements to the image.
The lora was also trained only with female characters in mind. If there is interest, I'm happy to add males to the dataset and add support for male characters in an update.
Notes:
From testing, the lora seems to also reduce the prevalence of artifacts in pale characters (that seem to be more susceptible in photo real models for some reason) when using Hi-res fix.
Duplicate eye pupil problem from V0.9 now fixed!
Description
Much more versatile and stronger. At high weights you can now create ghostly white skin!
FAQ
Comments (12)
"The lora was also trained only with female characters in mind. If there is interest, I'm happy to add males to the dataset and add support for male characters in an update."
Just throw in some Michael Jackson pics and GG
so long ok ...
but delete that negpromt you dont need and get much more better pics!
for me your lora is to strong! weight 0.3 and totaly different image
cheers for the feedback! What checkpoint are you using? I don't find significant changes until >~0.5 with the same seed?
@conr3000236 its my ... and maybe it s because the trigger (pale skin) doo already a bit.
so its always better to train loras for things that not be worded in the checkpoint ;)
@sevenof9247 good point! I'll try implement another trigger word unrecognized for checkpoints for next time.
After some testing I've found the lora does not need the trigger words and this seems to be what causes the high strength dependence. Therefor I've removed the trigger words as results are better without. Thanks for the help :)
I just tried to create a pale man and it's a pain, they're all tanned. It would be cool if you could update Lora for men as well
What is the trigger word ?
works without them but was trained to work with "pale" and "pale skin" to enhance it further
@JustBarry Thankyou for replying.
Can you train a Pony version please?
Thank you for this LoRA, good work.
it works fine 🧐
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.