NOT MINE
From SexGod1979
I made this pruned version in int8 convrot with turbo lora added after not finding it.
Took me 2h to find how to prune it correctly so at least I should share it :p
Description
FAQ
Comments (22)
Wow you managed to provide even less information what this is supposed to be than the original model card, good job...
It’s literally the turbo int8 convrot version of the pinkcherry finetune. What more info do you need?
@ornery54321 what is a pink cherry at all?
life is a mystery
Oh my God, Cherry is a model for generating porn, here it was cut down three times to run on something less powerful than the RTX 5090.you have a decent take but you could communicate your response in a nicer way thats just my opinion lol the model works good with int8 pruned
@Juicio Just hope it knows what a penis is..
@shanezwicker697 I am pretty sure feminists are polluting the Checkpoint-O-sphere with images of deformed penises. 99% sure.
@shanezwicker697 Use your own as reference image and get the best possible immersion just sayin'
@dillion1920 Nope. Not feminists at all. Actually, quite the opposite, it's all the ignorant straight dudes that are apparently the only ones making the base models, and most of the loras, and despite making them to distribute to millions of people, they feel the need to skew the datasets so hard toward 'female' that they barely know what a man even is.
@shanezwicker697 Will there be any downsides?
@extraltodeus I wish.. I had the unfortunate occurrence of having been genitally mutilated ('circumcised') as a defenceless infant. My body, someone else's choice, I guess.. So, mine is shite for reference. Even still, the reference image isn't the hard part, cuz without good loras, a badly trained model(IE, most of them) will turn dicks and balls into twats or futas..
Especially video models. Ref images work better for images, cuz it only has to pass 'frame1', and keep it to lookin like the ref, so, no next frame to guess what that cock looks like in a slightly different position than the first. Also, image models are relatively low resource enough to train, that there's many more variants, and plenty of gay/male ones by this point.
Whereas the resources required to train even a lora for Minimax, eh.. it'd be a stretch to pull it off on my card(16gb), and even in the best case scenario, I'd likely only be able to train using images, and it would take days, if not weeks, to do it; all while I'm currently in the process of developing an audio/video editor for Linux** (Currently working on polishing up a decently progressed copy to port over to android, so that I can continue with it on my phone, and eventually up it to the android/Samsung store once I've got a solid MVP).
** given the resources it'd take, even if I could do it, I wouldn't be able to use my PC to do anything else, for risk of OOM.
@Renessance to what?
Put in my /models/diffusion_models folder and it doesn't show up. Not sure what I'm doing wrong.
@shanezwicker697 You didnt address my point at all and the best thing your 12 iq brain could think off as p****?? LOL
@shanezwicker697anyways my point still stands and you're just a man child crying online
@Juicio My ass, you can eat it, all out. Lil piggy.
@shanezwicker697 my original comment got three votes more you got ratio'd have fun talking about your own ass alone
You're going the lord's work here, whoever the fuck you are. :D
Great job, ideally it would be nice to have the vesion without turbo Lora, so you can try different ones, which one you used? how many steps?
Which turbo lora was used with this?
Really useful model.
Dasiwa v1 4turbo is working but limited on quality.
Dasiva v2 is total junk crap same as Sparceref. Both creating messed distorted noise.
Singularity overtuned on basic portraits and not folllows prompt.
So no much competition here...
