I love seeing your sample images, it's really encouraging when you post your generated images in the gallery! Thank you <3
Was just messing around with training qwen to see what it could do - turns out it's pretty easy to train concepts!
If it wasn't clear this is an EDIT lora - no idea what it would do if you tried it with T2I
No idea what use this concept is for... but it definitely didn't know what it was before I trained it! It's not the most consistent, but it's kind of neat to play around with - qwen doesn't seem the best at figuring out what's behind obscured objects in a consistent way.
Description
FAQ
Comments (11)
Might be good for spooky ghost images with spooky season coming up thanks, I have never seen this concept in an edit model before.
Stupid sexy ghosts...
I bet this can be used to make some realistic looking slimegirls
Give it a try! I was just trying to teach a difficult but unique concept, it'd be great if it's actually useful for something
thanks for the suggestion, made a lora for realistic looking slimegirls
this is nice to make ghost. any better version planned?
it's hard to find realistic images to add to the dataset, so no better version was planned unless I can find some high quality inputs
it works just as well making clothes see-through
It does not.
OK, it's doing that for me so maybe it's just the qwen model doing it
@hamstornado or your prompt was more fancy than mine. I tried something real simple like "the clothes are see through" instead of the skin. I tried it with bf16 and uc-convrot model and was surprised how it refused to do anything. skin worked fine tho. :-)
it's not trained on see through clothes at all - might just be qwen can do it