SpongeBob!! trained off of a bunch of HD screenshots from the show and the movies.
"view from right side" and "view from left side" tags help determine the direction he's facing.
If you have any suggestions for a missing pose, action, prop, costume, or background, please leave a comment! I'll try to add it in the next update!
Description
this version has been trained on around 10x as much data as some of the first versions!
FAQ
Comments (7)
can we have more previews?
sure! i'm just waiting for this version to start being available on the site's generator 👍
SpongeBob clearly does not want to participate in the creation of NSFW content. Even the very generation of this character is very difficult for AI :)
Not that you asked, but I figured it might be interesting...
From my own training, I've found that NSFW data has to actually be present in the training data for it to show up. This is likely because without it, you're relying on your checkpoint to guess at what they look like with their pants off, while also simultaneously fighting the Lora's training data that has the clothes pretty much hardcoded into it.
Not that I'm genning Spongebob dicking people (haven't even downloaded the lora), but I've noticed this to be a common occurrence on Loras that use only source material as their training data to recreate the character as accurate and on-model as possible. I've also noticed the same happening when I train models the same way (using source material), but the issue goes away when I merge it with a secondary model I've trained on fanart (both SFW and NSFW).
@anon4324Â yeaah that sounds about right. there isn't a single other instance of clothing variation in this dataset, just the default pants, tie and shirt!
@Boongler Thank god for that too
He's asexual so only good that it doesnt generate NSFW of him XD





