For months, I've been trying to train a fucking LoRA for Hunyuan that would give me tits the size of Texas.
This was the result of being trained using fai.ai's trainer.
I unfortunately had to split this into two loras due to the image limit being low (seriously, 100 images???).
Use at a weight of 0.6-0.8.
Description
FAQ
Comments (18)
Texas called, they want their land back.
TELL EM I SAID COME AND TAKE IT
This must be really powerful then cause when i try to use it it says "something wrong happened" and im not sure is the site itself haha
The tensile strength of those bikini straps is seriously impressive.
That bikini was BUILT TENCENT TOUGH™
Awesome!
Looks amazing! Got any plans for a WAN version?
hmmmm wan...
I have to think about what I'll use to train WAN loras
Actually shocked. A video lora featuring hyper content? Every other video lora just focuses on realistic butt sizes. Heck that, give me cheeks and tits the size of Texas. California even!
Very great lora, dude. You knocked it with this one.
Thank you! I was getting tired of waiting for one after seeing how everyone that uses this thing would only train loras of celebs for deepfake slop.
Consider my addition to the Hunyuan community as stepping stone for change.
People need to wake up, this thing is very easy to train once you get the basics down.
@pork69 Incredibly glad to know someone's also annoyed of deepfake celebrity realism shlock. Of all the things to work with something as hugely potential as Hunyuan and you use it to see how your favorite celebrity would look naked or something. Ugh.
There's far more variety with Wan loras compared to Hunyuan, so I'm glad you're finding SOME way to bring that variety onto Hunyuan. I should probably do the same too, especially since video training now exists on Civit.
But, I don't know how good it is. The buzz goes well over to the thousands so I don't know if I want to keep my dataset small but not too small or large enough for it to at least be stable. It's a whole game of chance. At least I can use it without having to pay for a lousy subscription or needing 16-24 gigs of VRAM to train one.
@8747928hp
I tried the Civit trainer, too much for me sadly.
I remember trying to train the Gloria LoRA with it and it kept giving me failed results, so I went back to using the image-only trainer on Fal.ai, which pretty much does the same thing, only difference is I can't seem to know how many epochs it goes through.
Get the job done, though!
@pork69 Where can I find the image-only trainer on fal.ai? Can't find it anywhere.