/vt/ - Virtual Youtubers

>>49161195
you're in luck

>>49165544
https://civitai.com/models/48197/obese-girls-or-concept
be careful if you don't want ssbbw
if you want that better use the architect models, though I'm not up to date on that


>>49160120
back to cunny
with regular prompt + pubic hair :
https://files.catbox.moe/swsg4r.png
https://files.catbox.moe/o91lhr.png
https://files.catbox.moe/131312.png
https://files.catbox.moe/n8o6xz.png
https://files.catbox.moe/247lme.png
epic troll by wildcard https://files.catbox.moe/c3gymf.png

with kind slightly different prompt (forgot to change from the testing further in the post) :
https://files.catbox.moe/wu2ecw.png
https://files.catbox.moe/drykb0.png
https://files.catbox.moe/qmttz7.png
https://files.catbox.moe/bzsnfe.png
ass https://files.catbox.moe/79ca14.png


>>49160275
Hi again!
Talking about that, Seraziel isn't very compatible with chonkers, at least with a sufficient weight to see his style, so get something closer to the style I used for the fat and stinky orca (shylily), and some other lucky gens :
post's image
https://files.catbox.moe/m490z0.png
https://files.catbox.moe/za68fj.png
https://files.catbox.moe/zvk4ii.png
https://files.catbox.moe/4qt2hl.png


>>49160487
>>49160634
ALERT essay about the metaphysics of uncut cocks in AI
I can do LoRAs just fine, the problem is elsewhere. If you've been lurking on /d/ you'd see the average quality before the furry model mixes and recent artist loras showed up was abysmal, the cocks were flesh sticks, basically post-op ftm "cocks" even though futa loras weren't lacking, and I don't believe there wasn't a non-shit lora maker in that bunch. I believe it mostly comes down to NAI-derivatives having poisonned tags related to a lot of shit like cocks or feet.
I've tried a few of loras around this theme myself and on top of the concept itself being just badly trained, they all apply a shitty mixed style (and are also fried).
To properly use something like that you'd need to put it on top of an already working futa setup, but these are already very rigid in cock type (and style), meaning the model can't synthesize the two without a higher order understanding of this. That's because LoRAs are rigid themselves in how they work, and even block merging them won't do miracles : It's always better to be able to simply prompt a rock hard cock with foreskin than having to separately have 2 base futa loras (shit like futaveiny and some good futa artist lora) and then applying a foreskin lora on top because the other loras will fight against this one hard.
Hence furry models that were re-trained and probably understand what cock and balls are, in a way that's pretty flexible, and can be kinda ported to make slightly better futa models. The obvious flaw here is wrangling and lora incompatibility, but I've been able to make some good results, here are a few samples of pure gens :
More recent
https://files.catbox.moe/kxqg7n.png
https://files.catbox.moe/f07e81.png https://files.catbox.moe/nex6tv.png
Older :
https://files.catbox.moe/nqjkux.png
https://files.catbox.moe/pmqc3a.png
https://files.catbox.moe/4jnix8.png https://files.catbox.moe/b8kgwh.png

You can obviously see the hard wrangling in the prompt and in the image, but cock-wise I can't give up how good it CAN be (and not "WILL" unfortunately, since some loras dissolve into noise with this model, but if you get it working you're in for a good time). Some of that can be resolved by gradual img2img using a regular model, like I did for gura cock some time ago, but it's very time consuming :
https://files.catbox.moe/r56gyb.png
https://files.catbox.moe/dplvbz.png
https://files.catbox.moe/c7k1d6.png
But there are barely any groups wanting to do a new base model, and try getting VC money because AI is bad at cocks...

why are my posts becoming so long