Degenerates, all of you. How dare this be the most successful series of threads in years.

Just getting started? Lost? Grab a coffee and follow >>2 (Cross-thread)

This is the general catch all discussion, experiment, post, comment, and do whatever thread.

Old archived threads #3 >>1696 (Cross-thread) #4 >>2129 (Cross-thread) #5 >>9 (Cross-thread) #6 >>2539 (Cross-thread) #7 >>3352 (Cross-thread)

Request Anchor. Put your requests here

It's time we got an anchor, since these threads are starting to get busy.
Tag this post with your requests, to make them easier to be found and fulfilled.

>>4033
Requesting a catbox

>>4034
https://files.catbox.moe/6otlkq.png https://files.catbox.moe/kbqolp.png

>>4035
Thank you much

Someone on the forums really likes parroting the AI ART... LE BAD narrative.
>inb4
>dev uses it to improve the goofy 3d images with faces that look like that of a winnie the pooh
>NOOOOOO HOW DARE YOU
Never gets old.

Plugged some old request outputs into a new model.

>>4044

>>4045
And fat, sweaty Mio of course.

>>4046

>>4047

>>4048
How do I make artstyle lora's? Is there something I need to do for my images before training? And how would sd know to use the art style?

>>4049
I haven't actually trained a LORA yet (I want to but I don't think my hardware is powerful enough), but I think the way you do this is when you're tagging your images, you just need to mass tag all of them with a common tag that doesn't overlap with anything too common (ex. the artists name or 'art by name'), so you have a trigger word.

Here's a guide for more general info about LORA training: https://rentry.org/lora_train

Goblins!
https://files.catbox.moe/hh4hmk.png
https://files.catbox.moe/vcw3ux.png
https://files.catbox.moe/qcop6q.png
https://files.catbox.moe/brlt9e.png
https://files.catbox.moe/cptkb6.png
https://files.catbox.moe/yvbgjt.png

>>4053
Fat goblin sex!

more gobs

https://files.catbox.moe/i4gay6.png
https://files.catbox.moe/3plhfn.png
https://files.catbox.moe/tulx8y.png
https://files.catbox.moe/fmpixq.png
https://files.catbox.moe/qxwswt.png
https://files.catbox.moe/h8724j.png

>>4033

https://civitai.com/models/24294/defecation
https://civitai.com/models/29500/big-steaming-piles
https://civitai.com/models/33044/rear-view-hyperscat

Looks like civit took these down. Anyone save them and can reup?

Ouch, my dick exploded. :(

>>4058
>Literal scat fetishist
That's neat and all...but have you considered jumping off a bridge? You might void your bowels after you land!

>>4055
Fat goblins are S tier

>>4058
I have all of them except the defecation one (though I do have a different scat Lora that you didn't list). If you find that one, let me know.
https://files.catbox.moe/s9qp3f.zip

>>4060
complete waste of a post just to be rude and edgy. at least attach some fucking generated images so your posts have some worth

Some tests with LoRA mixes. Going to do another set with GAME and FINAL LoRA mixes.
If bbw model anon is there, might I recommend moving to a different base model? Perhaps something with Basil in it in order to try and get some good base anatomy in there? (Aurora might be a good pick) acidburn may also work well, given how many concepts it is trained on, but I think Anything v4.5 might be a hinderance rather than a help. (Too much NAI/AOM is leaking into the style that the model produces.) Although, a properly trained clip encoder on any new model will make merging 10 times easier, since you actually in theory could get good fats just from a clip encoder merge on a model not trained on fats.

>>4062
>white-knighting for an actual pile of shit.
>also has a scat fetish
Imagine complaining about post quality while turning the board into an actual shit hole.

>>4033
Requesting Rushia,Mikeneko, and Nazuna as big you can possibly do?

>>4049
In a normal concept or character lora you'd be fine to leave out some details, but in a style lora, every little thing needs to be tagged. The more tags the better, it helps it learn the style more (but you'll be fine with just using an autotagger)
If you want a trigger phrase give every picture the same tag that isn't used by anything else like 'art by [artist]' or '[artist] style' or something random like 'msks'
But the trigger phrase is optional and style loras will be fine without one. It'll activate either automatically or when you use the tags it was trained on eg 'wide hips' or '1girl'
Save checkpoints frequently, maybe every other epoch, because some styles will be learned faster than others and it might overbake faster than you think
The mint anon has a brief guide & settings that work well https://mega.nz/folder/sShj1LpS#Zh5ZxGO3uwK4QEaZ2X6z1w

>>4063
> might I recommend moving to a different base model? Perhaps something with Basil in it in order to try and get some good base anatomy in there?
Agreed. Anythingv4 looks great and all but it's starting to show its age
>acidburn, aurora
Was also eyeing aurora. Looks great and can do chuubas out of the box. Acidburn seems neat but it doesn't list or show any of the "many" concepts it was trained on, so I'm inclined to believe it's just a generic model, but with futa or horsecock or something since the maker is a /d/tard
>clip encoder
I'm still trying to figure out what the hell goes so wrong in training for the encoder to come out so bad. 65k even with a trained text encoder still had a fucked clip (anythingv4 only had a single position off)

>>4068
>since the maker is a /d/tard
I could have sworn it was someone from /sdg/ that made it.
>I'm still trying to figure out what the hell goes so wrong in training for the encoder to come out so bad. 65k even with a trained text encoder still had a fucked clip (anythingv4 only had a single position off)
Asking /sdg/ or reposing the question to /vtai/ might be your best bet.

>>4033
All I am requesting is nudity and ahegao :D

>>4052
catbox for 2 & 3?

Wide lion

>>4033
Requesting either Ema Skye and/or Maya Fey from Ace Attorney - preferably with a large belly heavy pear shape, thanks in advance

>>4048

Used everyone's favourite obesegirls lora with a monstergirl model and was not disappointed.

Mio and some Flare leftovers

>>4089
Catbox for those Flares please? The backfat spilling out in the last one is fantastic.

>>4093
>img2img of other fat art
>in the age of fine tuned fat models and model merges.
I mean sure. I guess.

>>4068
You are using Dreambooth to train the model right?
Are you also using regularization images as well?

>>4101
Love that girl in the black dress.
Can you please put this in catbox?

>>4101
I'm interested in the prompts / models for 2/3/4/5 as well, it might help me with the little story interpretation I'm working on for another anon.

>>4117
https://files.catbox.moe/czpij4.png

https://files.catbox.moe/c63z8r.png

https://files.catbox.moe/51tlei.png

https://files.catbox.moe/w4vu1e.png

Also, fun times happen when your controlnet doesn't quite line up with your prompts

>>4104
yes, and yes. I just use my gen folder as regularization

>>4136
can you box those last two?

>>4137
https://catbox.moe/c/2wemi2

>>4138
that works too. thanks

>>4136
That actually isn't what you should be doing for regularization. You should be taking your base model and generating the tags corresponding to each tagged image in your data set. Each tagged image should have a corresponding image generated from your base model for regularization.

>>4094
https://files.catbox.moe/wgzv5m.png
https://files.catbox.moe/5rhfgc.png

Kiara after the KFP company picnic. In unrelated news, the have a few new positions open

At least it wasn't a repeat of the chrostmas parties

>>4153
Thank you.

>>4055
Thank you

>>3656 (Cross-thread)
> Mumei as an average American at Walmart
She strikes me as the type to go shopping while hungry, bring home her weight in snacks and in-store deli food, and then get annoyed at how much junk she eats

> Fauna eating bread because veganism
'Wait, bread makes you fat?'

>>4152
ahh, alright, thanks for explaining. It's surprisingly difficult to get a straight answer for some of this shit.

>>4155
WEW catbox for 5?

>>4161
>WEW catbox for 5?
https://files.catbox.moe/8wcjd3.png
most of these are hidden in vtai with normal sized Kiara as the post image

>>4161
https://files.catbox.moe/g5yd3s.png
and I cannot count

>>4162 >>4163
> most of these are hidden in vtai with normal sized Kiara as the post image
Evil, I love it.
Thanks, as well

I don't know what compels me to keep making big Mio, but I don't plan to stop either.

>>4161
Could you box the Mumeis please?

>>4165
It's the /vtai/ staple.

Different anon trying to make their own Lora here.

Got 120 pictures with their respective txt file, made a +2 GB file that doesn't seem to work. Didn't do it on Google Colab.

>>4167
>2gb file, not on colab
I'm assuming you are training with the dreambooth tab in SD. Don't
Training within SD is really bad. It only trains checkpoints, throws NaNexceptions often, requires a hyper specific version of a half dozen dependencies (and will just fuck your training instead of warning you), and breaks after nearly every webui update. Don't bother

Use the kohya gui https://github.com/bmaltais/kohya_ss and if you have trouble setting it up, you can also use the training script https://github.com/derrian-distro/LoRA_Easy_Training_Scripts . They're the same thing, one is just a gui like webui and the other is a script
You could also use the colab, loras train fast enough that you're unlikely to be timed out with 120 pics, and even if you are, you can just resume training from an intermediary save

https://rentry.org/59xed3 this is a pretty comprehensive guide
https://www.youtube.com/watch?v=70H03cv57-o also a good one

>>4166
https://catbox.moe/c/cnnzzz

here's a couple i messed with, most with that bariatric lora!

>>4168
Thank you.

>>4161
I should mention they should be gened at the same resolution as your training resolution

Is there a stretchmarks lora?

Huge Silvervales, for your viewing pleasure

>>4168
Thank you

Finally made a Moxxie Lora, had to spoil it. There's no fat male thread and don't have enough material to justify making one at the moment.

>>4192
catboxes or album for the whole lot please?

>>4194
https://files.catbox.moe/b09z3b.zip

Anatomy just kind of stops trying after a certain size

>>4195
thank you!

not entirely fat, but was made using the gods amongst men version and inpainting in that giantess lora from civitai

Been learning a decent amount, having fun with all of this. Been inspired by the posters here. Can be kinda frustrating at times to get good stuff, but it can be really good sometimes.

>>4200

Giantess stuff is real fun too, easier to get than fats too for obvious reasons. Only attaching one here because this is a BBW board and it's not fats, but wanted to share.

>>4204
These new threads have been a success, the more the {{{bigger}}} we can prompt.

>>4033
saw some really good nina kosaka on the other threads. More of that. any way.

>> 4204

mind catboxing the 5th image?

>>4204

mind catboxing the 5th image?

>>4207
I love Nina, do you want BBW, Giant bbw, pregnant, chibi chubby, chibi with huge hips/thighs, or a mix?

>>4226
>
Love these, If i had to pick, probably bbw with extra curves. I prefer seeing more weight up top.

>>4237
These models are really focused on hips but I got some good curves

>>4211
https://files.catbox.moe/xblz43.png
This is the original image. I cleaned it up a little in CSP and swapped the face out because I couldn't get it to inpaint a good face. Just stole another face from another generation and threw it on there, and then colored in the multiple belly buttons and blended them together.

>>4226
>>4250
Different anon, but if I could request some giant BBW Nina, that would be great. These look amazing.

>>4200
Holy shit that's good. Need more chubster giantess'

Maybe they're good, maybe they're bad, I dunno

>>4254 They're great.

Elira plumpening. She's really putting those jeans through hell

>>4193
Good job on the train, anon. Keep us tuned if you make any more ;)

>>4204
Catbox for 4?

>>4256 I kind of want to take her back to my place for a few drinks.

>>4251
>>4253
Perfect time for me to try out a giantess lora as well

>>4258 I was thinking ice cold....... iced tea

Still request. Huge Haachama and Okayu

>>4256
https://files.catbox.moe/uvgvw7.png

>>4256
Catboxes for this sequence please? Great work

>>4254
>Maybe they're good, maybe they're bad
>I dunno
Some help you are! Is there anything? Anythingdo know?

Random blob fox. Result of using a BBW model with a BBW lora

>>4033
Maybe some super bottom heavy fat Hu Taos?

>>4260

>>4031 (OP)
Holy shit, catbox for 4 please?

>>4062
>https://files.catbox.moe/s9qp3f.zip

Fortunately for you shit man, I happen to have that exact one. You saved me.

https://files.catbox.moe/09emsu.safetensors

You'll want to rename it to defecation_v1.safetensors

Okay now to never open this website again, I hope you guys live happily ever after with your whales. You have my infinite gratitude and respect.

Dragon(nite) waifu concept. Let me know if you want a new thread for monstergirl, but I'd probably be the only one using it, lol

More Mio, totally worth the time

Alright, mergeanon here. I have uploaded 5 new experimental models to my experiment repo. Final-LoRA mix is weird, has the same pastel problems as Final. The basil mixes are a mixed bag. They do not respond very well to certain tokens even when weighted to like 1.5, (specifically eye color, and clothing color), which leads me to suspect that there is something wrong with the clip encoder, or somethings fucked beyond repair in the UNET due to the merge. Outer Heaven is Basil mix with the LoRA Mix.
(Try bbw_GAMEMix, as that's the one that I think works the best.)
Try them out, and don't forget to leave some feedback.
https://huggingface.co/mariaWitch/ExperimentalBiggerMerges/tree/main/Testing

>>4299
Keep it up, will try them out when I have time.

Set up webui last night and start having a go at making some wide kallens with a lora from civitai. These threads got me on the right track to generating something that isn't shit, but i'm still trying to wrap my head around it

>>4299
I'm a big fan, thanks for all your hard work. I'll be DLing all of these to try later tonight

>>4302
A lot of trial and error, just try to tinker with parameters, you can use multipliers to change weights on words

>>4299
Are ckpt lora and safetensors models? Little new to all this.

>>4306
Yes, they are the same.
Waiting for huggingface to actually scan them so that people don't have to be paranoid. But if people are super paranoid just use a pickle scanner.

>>4305
Yeah I’m noticing there’s tags that don’t do shit depending on the model/lora, so it’s a lot of tinkering. It’s a fun thing to have going in the background though.

Is inpaint upload and mask a reliable way to touch up faces/hands/general errors? I’ve fiddled around with img2img but I haven’t had good result from that so far.

>>4306
Sorry, let me rephrase. Both are used for both. You can tell a LoRA from a model based on size.
A model is at least 2 gigs, a LoRA should always be under a gig.

>>4311 I love your art. awesome post

>>4256
i am also requesting catboxes or an album

>>4308
Had success looking at how to tag in gelbooru, Waifu Diffusion tagger for making artist Loras is amazing. It's weird to put into words in describing art styles. Just a sligar Lora, but it's annoying to deal with colored limbs that are hard to get rid off in negative prompts.

ive been figuring out how to use these models and things without making abomination flesh piles and theyre starting to turn out pretty well now

>>4316
I'd have her sit on me. Try using highres fix to get some more detail

shit i dont remember what prompts i used i cant even recreate it

>>4323 Feed it back into the AI... replicate. Or commission a professional artist.

>>4299
Some Ame using basilmixfp16.safetensors, I like it so far
>>4323
That's where the PNG Info tab comes in handy on the webUI.

the png info just shows me what i put in for the inpaint prompt so its missing like half of it. i dont have the money to spare to commission an artist, and its not like im selling what i generate or anything

>>4325
Catbox for the last one please?

>>4326 The basis of the technology is replication. You should be able to do so minimally. If it needs more than 1 for training then make many copies of the same image and feed it many copies of the same photo.

>>4327
https://files.catbox.moe/8tpqt5.png

Amazing how AI art seems to keep getting better.

On the quest for plump Gura, also the "beach ball" tag seems to have made Chloe an inflatable
>>4330
That's what happens when there's a massive open source project that makes porn

>>4332
box for 3 and 4?

>>4330
>>4333
https://files.catbox.moe/2cvmgr.png
https://files.catbox.moe/wvuckw.png

>>4332
Gave me ideas to tinker with...

Iofi's bakery with plenty of cake
https://files.catbox.moe/mhse59.zip

>>4335
Really good, but I can't help but want to push her over if she's squatting in heels like that

>>4332 Dangerously loli. I mean, cheesy.

>>4334
thanks

Should we make a separate thread for vtubers?

>>4339
No? That's where most of the content is coming from.

>>4339
The furry thread is dead, better to make an effort to keep this thread active.

>>4339
Fuck segregation.
Not aimed at you, but if anyone sees something they dont like just hide it. Its in the no parking looking symbol then hide post

>>4256 >>4257
It didn't take much convincing to bring Elira home. It did, however, take a whole lot of pushing, wobbling, squeezing, and a little bit of fondling even though she could've just went in sideways... She didn't seem to mind, though.
A 'few drinks' also pretty quickly turned into 'everything in fridge, pantry, and cupboards'
Good luck getting her through the door now.

are these short stories cringe? let me know
>>4294 >>4339
>monster girl thread
we post kemonos here on the daily, as well as slime girls and lamias, your posts are just fine here imo. We also have the furry thread. Not much point to a monster girl general
> vtuber thread
I think the chuubas are also fine here unless we get a surge of new users & non vtuber posts at some point. Splitting the threads apart now will just make them shittier

>>4261
tyvm!

>>4264 >>4314
glad you like it :) here's a catbox album for the whole set
https://catbox.moe/c/qro6nm

>>4282
https://files.catbox.moe/1cd8ca.png

>>4349
>cringe
No, not in my opinion anyway. There was a guy on /vtai/ who used to make little stories with his generations too.

>>4350
Thanks for the album as well.

>>4349
Nah, they're fine. Agree, that splintering the threads at this point is a terrible idea. AI art is quite new, interesting years ahead. Would be happy to see if there's better automatic taggers for AI porn than Waifu Diffusion tagger.

>>4336
To fall on my face, hot

>>4033
could someone do one with pink hair, freckles, and a massive ass in tight shorts?

>>4357
Good enough?

>>4358
yep. thank you!

I'll just dump everything I got in here for ya'll

https://mega.nz/folder/GB9mmBYa#NEJ9iWmGOEbIdrjoRlDcyQ

>>4362
Thank you!

>>4033
Humbly requesting a preg/expansion sequence of [insert VTuber here] as they go from their canon figure to hyperpreg and hyper tits, with appropriate amounts of thickening elsewhere. Look up CoffeeSlice's Samus/Rosalina/Zelda sequences for the general idea. (Also because they're fucking amazing)
Options are Sora, Mel, Fauna, and Anya . Too hard to decide, but I think Sora or Mel are near the top. something about them makes for especially hot em{{biggen}}ing.
>>4226
NinaAnon, if you're still offering, I'll take some heavily pregnant Nina, especially if you can get her tits near the same size as the belly.

>>4364
That's going to take effort because that involves making a Lora on CoffeeSlice's pregnancy sequences.

>>4364
First of all, I have some preg Nina. I'll work on getting breasts as big as the belly, I have a suspicion that SD confuses the two sometimes. Second, I'm very familiar with coffee slice. I'll do a Mel sequence when I have time

>>4367
And a side note, I'm trying to get her look more pregnant than fat, but these models are tuned for bbw

>>4368
(pregnant belly:0.5), (pregnant belly:1) and (pregnant belly:1.6) seem to work without any Loras on bbw_hll4b2. For a first impression, this belly tag has a different form from huge belly outputs.

>>4366
Sorry, I didn't mean "give me something in this artist's exact style", I just meant to give some context to the request. I don't think there's an agreed-upon term or tag for it (which makes finding that kind of content with any specificity a boner-shrinking headache), and it's admittedly a work-intensive request (you are all kings BTW), so I didn't want to make someone waste the effort it would require on something that ended up being only like, 40% appreciated. I mean, even when I've done requests that didn't get filled quite right, I still appreciate the hell out of it.
>>4367
>>4368
My man that's some good Ninaur right there. No longer just an adoptive mom, but a biological one too! And yeah it's a thin line to tread, but a more uniform roundness definitely sells better as preggo than rippling flab does.

>>4371
Mel Series, not perfect but I managed to go from normal to big to huge.
https://files.catbox.moe/l542r9.zip

>>4372
I haven't even opened the zip yet but just by the previews you fucking nailed it, pic #3 is so goddamn tasty

>>4371
No problem, it's just hard to visualize that request at this early point in AI prompting.

>>4376
All good, my dude.

Something about the way Mel acts just makes her feel like she'd literally let you do anything you want to her and she'd both just be sweetly happy as well as get off from it. A true bimbo, not in stereotypical appearance, but in purpose.

?

could someone do qiyana from league please?

>>4381
The Lora changed the artstyle, nice request nonetheless

>>4381
Damn, they're amazing. More gens here

>>4357
could someone do something like a back view of this?

>>4386

>>4381
AAAAAA
Want her to sit on my face

>>4388
thank you so much!

>>4372
I wasn't the requester but thank you for this.

As the original one who requested Nina, the amount of additional Nina gens is a W. Thank you for the food. And the fellow nina appreciates.

>>4033
if anyone's up for doing furry (and male) stuff, could someone do a back view of a pink cat with a huge ass grinning at the viewer?

>>4399
This is still BBWai bro. What does the W stand for?

This is why a general thread is a shitty idea.

>>4402
and there's a bhm thread

>>4404
should i just request it there instead?

>>4399
Want to do it but male fat furs are hard to prompt on these BBW models.

>>4406
female would be fine

>>4399
As much as I enjoy chubby femboys, I haven't even touched furry AI stuff yet. I'll work on it and post in the furry thread

>>4409
got it!

>>4405
This is a catch all AI board. If you want to make male stuff you would probably have better luck making a male thread but this is a general so you could do it here to. Ignore the faggots and be gay all you want. Cry babies really need to learn how to hide posts they don't like.

>>4411
>Ignore the faggots and be gay all you want.

That word has lost all meaning to retards like you, hasn't it?

>>4226
>>4033
any chance you could supersize this character? (or one vaguely similar to it)
belly grabs appreciated if at all possible.

>>4418
Once I get home will give it a shot.

Made a chubby Kronii and the bed is breaking under her

>>4418
I gave it a whirl and I like the results. Might even make her bigger.

>>4420
How did you get it actually breaking in the first one? Catbox please?

>>4422
Just happened by chance. AI must have picked it up from whatever the model was trained on
https://files.catbox.moe/z6d6n8.png
https://files.catbox.moe/kb2adh.png
https://files.catbox.moe/eeagqx.png

Here's more of this big fox girl too, still working on getting her outfit right

>>4421
>>4423
well ill be damned. These do look real good. I certainly wont complain about bigger or bonuses. Thanks!

>>4423
Based model, I'll have to give a try.

>>4424
I did use the img2img feature and got this cutie, but the rest are more eyeballing it and getting the prompt to generate a similar enough result. Still a nice looking character to work with, and making her bigger is my pleasure. Here's some belly squeezes

>>4426
Thanks for supersizing the fox. Loving the second and fifth ones. Good batch. Squeezes and grabs really are they best aren't they?

>>4428
Squeezes and grabs are fantastic. Here's all the batches I did so you can replicated them if you want
https://files.catbox.moe/2h7pma.zip

>>4033
Can I request a fat ssbbw Satori Komeiji from Touhou project?

I'm a fan of how active it's been getting

>>4278
> super bottom heavy fat Hu Taos
Patrician taste. how's this?

>>4399
I'll try it out in a little

>>4033
Requesting a bottom heavy inquisitor girl dressed like the Seventh Sister from Star Wars

>>4431
Can you catbox the first and third one?

>>4431
Those look great, yeah. The fourth one with her being almost completely flat is a great detail too, thanks!

is the 'gods amongst men' BBW model still the popular one, or has it fallen to the wayside for other stuff?

>>4436
Still pretty fantastic, especially if you want bigger butts and thighs

GAME does really big really well. The new GAMEmixfp16 is also available (pic related). I personally like the bbw_hll3 basil-mixfp16

>>4436
That's what I still use, but I haven't tried out the new experimental merges just yet

First ever time using image gen. Wanted to see if I could get it to do monstergirls. Took an absolute shitload of inpainting, but I kinda got it to work with a lamia

>>4410
Sorry bud, couldn't get a result I liked. I'll work on it and post in the furry thread, we all know they need more activity

>>4349
No, the stories are good and your images of this girl and the sequence they form are god-tier. Please, for the love of all things fat and sexy, give us more of this girl on her growth journey, cause holy fuck my guy. Shit's good.

>>4441
Try this. There's a lora version that plays fairly well with bbw models or use the monstergirl model with a bbw lora, I've done both

>>4441
https://huggingface.co/scriche/AIOMonsterGirls/tree/main
Sorry, this

>>4445
Thanks, I'll give it a try. Still very new, and experimenting with all the settings lol

Curious if anyone knows how, any clue on how to fix this image? I love the way it turned out but it's obviously pixelated and has some issues. Generated it again with the same seed but it still came out pixelated, my computer also had a stroke trying to generate it.

https://files.catbox.moe/ygj9bt.png

Quick example of the aiomonstergirl model with obesegirls concept lora

>>4447
> Denoising strength: 0.4, Hires upscale: 2, Hires upscaler: Latent (nearest-exact);
this is the issue. use latent bicubic antialiased with a higher denoise. for GAME i like 0.52

>>4449
That did the trick, thanks! Appreciate it a lot.

>>4431
I was also gonna ask for a box of 2 and 3. Might as well just throw in 4 as well.

>>4455 >>4433
https://files.catbox.moe/0qp31e.png https://files.catbox.moe/bq96ti.png https://files.catbox.moe/w227jn.png https://files.catbox.moe/ab6ypi.png

>>4351 >>4353 >>4443
kind words, noted!

>>4399
>>4456 (Cross-thread)

>>4460
https://files.catbox.moe/bs0n4i.png
Mad kudos to you, honestly. One thing I've been trying to crack from your end is the texture you're getting in your pictures. I really really like how everything gets rendered, but I can't get it to work on my end, even with a direct import of your pictures. Here's my result of the hose feeding one, can you see where I'm going wrong?

>>4432
Figuring out how to prompt this, might be an interesting Lora to make. Give me some time, maybe repost your request with more references.
>>4460
Thanks for the Moxxie compliment, still considering pruning tags before making the Lora public. Your proactive posts are encouraging to the whole board.

>>4460
Thanks for the links.

Ok, so I found these on thread number 6, does anyone know what prompt was used for this?

Found a few tubby lolis

>>4432
>>4467
Here are the results, might upload them to Civitai, getting the pics to train were a pain. Happy with the results, took me 15 images each to train the Inquisition and Seventh Sister Loras. These were made using 0.6 weights

>>4494
>>4432
Immobile Star Wars

>>4494
> Do you have a link to the lora?

hi

>>4496
Let me upload it to Civitai with example and a description.

>>4033
Requesting a very heavy Yiga Clan (BOTW) girl dressed like Master Kohga

>>4496
Busy rn, so they're here to grab. Trigger words are "inquisitorsw" and "seventhsister", each have their own Lora.
https://files.catbox.moe/yuay1l.safetensors
https://files.catbox.moe/jg1u46.safetensors
>>4499
Maybe later

>>4500
>>4496
Forgot to add the images with tags.
https://files.catbox.moe/0jh0h1.png
https://files.catbox.moe/gqtfvc.png
https://files.catbox.moe/cfl0kc.png
https://files.catbox.moe/1agn2y.png
https://files.catbox.moe/61umie.png
https://files.catbox.moe/jhuttb.png

Obese girls concept Lora
https://civitai.com/models/48197/obese-girls-or-concept

>>4498
Can you catbox these please

>>4500
which one's which?

>>4505
Fuck, didn't know catbox overwrote the names. Give me a moment.
>>4504
Yes
https://files.catbox.moe/14vkk0.png
https://files.catbox.moe/phcj1b.png
https://files.catbox.moe/gkfehj.png
https://files.catbox.moe/0anuma.png
https://files.catbox.moe/mtivoi.png
https://files.catbox.moe/r8uyke.png

>>4505
Rename this file to seventhsister
https://files.catbox.moe/jg1u46.safetensors
Then this to inquisitorsw
https://files.catbox.moe/yuay1l.safetensors

>>4349
How about a continuation with Elira laying on her belly with her ass towering above her? Perhaps the jeans are all but rags at this point.

>>4476
Haven't quite figured it out yet but there are monstergirl loras, I posted one earlier in this very thread

Oh this one works way better, lol
https://civitai.com/models/15112/slime-girls

>>4509
>>4460
The "pushing her through the door" was also pretty amazing. I don't know if the AI can do it with any degree of skill, but showing her completely unable to shove her way through the door, or better yet having to carried outside by people/heavy equipment through a knocked-out wall would be insanely awesome.

I made some fat lamias last night

>>4532
Catboxes for 2 and 4 please?

>>4533
https://files.catbox.moe/5w3jzh.png
https://files.catbox.moe/s43tb6.png

Cooking up something new here for all you anons.
I'm moving away from my previous merge recipes and am going to be trying something VERY different with the next experimental model merge.

(instead of trying to build up and fix what are ultimately flawed models, due to them being "concept" models, I am going to try merging them into an already existing model. Base model is incredibly important here, so I am going to go with a proven model, which also has basil in it, to hopefully strengthen the anatomical accuracy/resiliency against the fat models, which have a tendency to devolve pretty horribly in certain instances. An example gen with the model I am going to merge into is provided as the second picture. The goal is to keep that overall quality while just adding fats.)

I will be posting In Progress photos as I work through the merge, as I will end up manually scoring nearly 2000(!) images in order to complete this process.

>>4537
Go for it, we will be here for it.

>>4537 This post's hilarious. Sounds like my neighbor wrote it.

>>4537
godspeed anon .o7

>>4379 (Cross-thread)
Her royal Thighness. Grovel, and maybe you can have a turn as her throne pillow.

>>4462
> I can't get the texture you're getting in your pictures
try the latent upscalers, especially bicubic antialiased on .50-.55 denoise. Latent is all around better at details, clothing, texture, skin texture. The only reason to not use latent is if something messes up, which might happen with low denoise, some loras, latent couple, controlnet, or intense prompting
Inpaint and img2img will also go a LONG way in fixing faces, errors, adding details, and making sequences. For reference, the hu tao bed pics were both the same picture, this is the original and edit: https://files.catbox.moe/y8a11c.png https://files.catbox.moe/xmux9w.png
Another example is https://files.catbox.moe/j6wndo.png which is the original for 3&4 of >>4349
there's a brief guide here that still holds up >>7 (Cross-thread)

>>4537
> manually scoring nearly 2000(!) images
jesus christ

>>4476
https://catbox.moe/c/7xzl1r

Some chubby callis plus a rando

>>4534
Thank you.

>>4537
Some final updates before I start doing the merge.
I am locking Base_alpha, which is the value which controls how the text encoder is merged, to between 0 and 0.25 as I have significant concerns about the fat models text encoders, and don't want to dilute the models ability to respond to prompts too much. My hope is that we just modify the blocks that light up when fat related tokens are put in, so that we get fats, without losing the ability to properly prompt chuubas, other girls, or good backgrounds.

>>4546
Catbox or model name?

had two false starts where changes I made caused errors during the scoring process. Should be sorted now.
Here are the first gen from both failed merge attempts.
I didn't get a chance to score. So let me know how you would have scored these two images. Note: these have not been through hires fix and only went through 20 steps of DPM++2m karras at 7.5 cfg, so you'll need to make some guesses regarding potential final quality.
(I have adjusted the prompt to include blue eyes after two consecutive red eye gens for haato.)

>>4555
https://files.catbox.moe/ccez47.png
https://files.catbox.moe/z8ixbu.png
https://files.catbox.moe/qz1xsh.png
https://files.catbox.moe/3aikds.png
Along with some chubby Wamys

>>4559
Crap, the model.
Here's the link:
https://civitai.com/models/7371/rev-animated

When I said the fats models cause anatomy to devolve, I meant it. This is so bad, that I am tempted to rate it below mosiac tier in order to try and send a message.

>banned for full clothed chuubas being "NSFW"
okay lmao

Welp guess I'm stuck here for a few days because some /vtai/ janny decided that my full clothed chubby chuubas were some how nsfw, when full bikinis and cameltoe aren't deleted.

>>4568
> few DAYS
Why are the mods such fags? What do they get out of ruining perfectly good threads?

>>4568
>banned for chubby chuuba because janny felt ashamed busting dust to it
Sorry anon, you were too strong

>>4567
>>4568
That sucks, more should move here.

>>4568
At least you're in good company here

Here are some selections from the 6.25 bucket. We are about half way through the initial test gens.

>>4568
It’s bad. All those haters of AI and members of artists’s lobby of anti-AI are very dangerous and emotional so they use those illegal methods of war against us. It’s very sad

Enna was concerned about being underweight, so I helped her along a bit

>>4577
This is great, keep it up

>>4580
Catbox for 3 and 4 please? Very cute by the way

>>4585
https://files.catbox.moe/ebpf0j.png
https://files.catbox.moe/rm73m9.png

How much money do you make on your arts?

>>4587
The dopamine I get from making others happy is more than enough payment. Also I hate paying taxes

>>4568
BBW artist LoRA anon here. I'm not banned on /vtai/ but getting really sick of the jannie bullshit. Fucking up very cooperative and pleasant threads. /vtai/ posters would be better served here.

Glad to see this place has lots of activity. I really should stay around more often.

>>4587
None.
I'm not opposed to the idea of making money off ai art, though.
"B-b-but you're just typing in a box and a picture comes out!" if it were really that easy there wouldn't be so much shit floating around. There's actual skill and time that goes into making something halfway decent.
Obviously not as much as if you were to actually draw something, but I don't think it's unreasonable to pay someone for their time and skill they developed over time.
Doubly so if they inpaint or do any image editing, and triply so if they actually make their own models and loras.

I think the real issue is that you probably couldn't profit off it even if you tried. Unless you also do a bunch of other fetishes and/or loli, fat won't get you anywhere, it's too niche.

>>4587
None. Making no effort money off the backs of other people's work is scum sucker behavior. I only make AI stuff for fun.

>>4033
Requesting a blob/SSBBW Herta from Honkai Star Rail. Been trying to generate some on my own but they haven't turned out good, so I'd like to see if someone else could do something. Here's a Lora: https://civitai.com/models/54312/herta-honkai-star-rail-doll-joints-enhanced

>>4589
Good to see you here.

>>4594
Maybe the Lora wasn't done properly, automatic tagging without pruning causes these problems.

>>4596
Works decently enough for nonfats. I'm really into giantess so I made a couple of images with a giantess lora and the results turned out fine. Don't know enough about this stuff to know what that means, if it's model incompatibility or what since I swapped to anything-v4.0 for GTS instead of Gods Amongst Men, but figured I'd mention it.

Spoilered images as reference to show what was output

>>4598
That's the problem, the tagging may be wrong. Maybe it's using too many wrong tags in making fat art. Maybe if someone bases it off a bbw model and prunes obvious red flags in tagging, it may work properly.

>>4528 (Cross-thread)
>>4033
could someone do this if they're okay with it?

almost 500 images down.
I am mentally exhausted from this process. I didn't expect it to take so much out of me.
Here are the 6 images from the 6.75 bucket.
I have one last testing iteration to go, and I am probably going to leave the next 10 for tomorrow, I am having trouble actually keeping my rating standards intact while doing this.

>>4602
Going through hundreds and hundreds of images to rate them is hard work! You should definitely take breaks.

Anyone try loras that were meant for skinny models with fat models? The Jack-O-Pose sorta works... sorta lol. Same prompt, with and without the Jack-O-Pose.

You know who I bet not many people think of when thinking of bbw vtubers? Reimu. I think she's cute, so I made her a cute blob
>>4594
I'll give it a shot

Inspired by Bimbo Mell anon, I tried to make some empty-head-full-belly ones

>>4594
How does this look? I don't know much about the game or character

Experimenting with a quick-and-dirty wg sequence approach: X-ploting the emphasis of a group of fattening tags. Not the most coherent but it gets the job done.

>>4610
This is really cool. Can you post a zip of the individual frames too?

>>4611
sure, with bonus sequence. That's it for me tonight though. I might go through my series on the ancient temple theme. I have hundreds of known good seeds to do this to.

https://files.catbox.moe/etzjfn.7z
https://files.catbox.moe/d1drdq.7z

>>4613
Thank you very much! If you do decide to do more of these from the temple, requesting the biggest Fauna and Elira get the treatment. Have a good night!

>>4299
i ran some xy plots for the experimental models. I noticed a few things:

Size doesn't seem to be an issue and the new merges can go about as big as before, from what I can tell
Basilmix has VERY nice details, the best of any of the models. But it does, indeed, completely ignore some parts of your prompt (enna was supposed to have 'white wings' and that somehow got mistaken for twintails?)
Across the board, intermediate is identical to outer heaven. I would consider it objectively inferior to outer heaven because I like OH's name better :)
Every model has poor tail placement and can't do animal ears from behind. inpaint or loras are still necessary for fixing them (but this isn't unique to the bbw models, so whatever)
The lora mix has the same issue as last time where it adds a lot of noise to the image, often to the detriment of quality, and also doing weird things to shadows and skin color (why is bae covered in paint and confetti? why does irys look like a vampire?),
Interestingly, the above isn't a problem with OH despite OH using the lora mix. OH is overall a nice mix between the stylization of GAMEmix and the realism of basilmix

It's worth noting that I didn't test lora compatibility, original outfits, many poses/sizes, or expressions, among other things

Overall I don't think you can go wrong with any of the models. Even the older ones are really damn good. The only losers here are intermediate and loramix. Also, hllb4 did really well- why did we say it sucked, again?
Personally I'm going to give GAMEmix and basilmix a spin and see if I can pick a favorite. I might also just stay on GAME

I have comparisons but they're too big for bbwchan, even when compressed to jpg, check them out here:
https://mega.nz/folder/6dVRhbYb#B5EOtVmpl4ppXfQVVa_GNg

>>4622
link to GAMEmix and basilmix?

>>4625
>>4299
> https://huggingface.co/mariaWitch/ExperimentalBiggerMerges/tree/main/Testing

>>4627
thanks

>>4605
god damn, catbox for 3?
>>4606
could I also get a catbox for 1?
>>4608
and 3?

>>4499
Oooh, yes I want to see that too

>>4629
https://files.catbox.moe/3p687b.png
https://files.catbox.moe/uoft6s.png

>>4608
Those look good, yeah! Thanks for this.

>>4622
I've been using Outer Heaven today and agree, the name is pretty nifty. Here are some results

More Outer Heaven

>>4639
Catbox for the first one please?

>>4592
I can't agree. We all make money from other people's work, artists always make it. So it can't be bad, if you make money from other people

>>4602
I am awake and refreshed. I finished bucketing the remaining images that I didn't do before I fell asleep last night. Here are the first 500 images.
https://pixeldrain.com/u/rSAD5jU6
Break down of the buckets:
3+: Major Anatomical issues, Significantly Deviated from prompt, Burning, blurry, or Very bad NAI fats.
4+: Minor Anatomical Issues, Minor Deviations from prompt, good non fat gens live in the upper end of the bucket. NAI fats, or bad proportions. (2girls being genned unprompted also land here and the previous bucket)
5+: Great Non-Fats, Good Fats (No NAI), Good Anatomy/Proportions, slight deviation from prompt, good background and pose, may also contain images that would look good if hires fix was used
6+: Exceptional Non-Fats, Great Fats, slight clothing color deviation from prompt, Very good anatomy, Great Background or Pose, Creative gens also end up here, as well as images that I think would look great with hires
NOTE: Due to the nature of this process there are a mixture of different images, while almost all of the prompts were fat focused, not all of the images ended up being fat focused. Which is just the way these things work.

https://civitai.com/images/663639?postId=183509

>>4673
I'm sorry, can you remind me again of which models you're trying to merge again?

>>4677
This is really impressive. I didn't have a high opinion of negative embeddings in months past, but maybe they've gotten better?

>>4068
Mint / artist LoRA anon here, catching up with the thread. I'd just like to let everyone else know that anyone thinks one of my LoRAs is overbaked, I can pretty easily upload a prior checkpoint, or simply retrain it with less steps. I usually just train ~8000 steps, make sure that the final checkpoint looks good, and then upload it. I don't take the time to validate the other (usually 15) checkpoints to check which is the "most optimal".

Also, if someone's favorite artist (or simply an artist you think is high quality) is missing, I might already have the dataset for to train a LoRA. I think I've covered a lot of different styles though.

Furthermore, I train my LoRAs from NovelAI, because I had come to the conclusion that it was a good base for use in different models, and the results looked fine to me. Does anyone do anything differently?

>>4299
Mergeanon, my current mainline model is Holocreamsicle (hll4e3 and AOM3) mixed with 40% BiggerGirls Dream IRL. Do you think my usage of AOM3 is a fuck up? I think I'll run a comparison with OuterHeaven.

> You have been banned from /vt/ for posting >>48618797 (Dead), a violation of Rule 1: Off-topic; All images and discussion should pertain to virtual YouTubers ("VTubers").
> Tubby tuber is off topic
> Hyper muscle, futa bulge, feet, loli, cameltoe, aerolae, pokies all untouched
the usual. anyways i got some moomers

>>4631
thanks very much!

>>4686
that's a very kind offer and I might take you up on it sometime
> I train my LoRAs from NovelAI, Does anyone do anything differently?
I think you're good. i tried training loras and i used bbwGAME, then anythingv4 and cocoa because they're the base models, and then novelai. The novelai training was far and away the best, which I really didn't expect- you'd think that a lora trained using your model and used on your model would be the best, but I guess not

>>4687
Janny is a gay low-test seanigger, apparently

I never stated what the base model was, but the base model is Aurora. And a custom mix of bbw_30k_step and BGv3.
I am merging using Bayesian Merger's Manual merge mode. I have gotten through half of the second part of the merge. I am currently taking a break since it's very tiring to rate all of these images.

>>4689
Forgot to quote >>4686

>>4687
>you'd think that a lora trained using your model and used on your model would be the best, but I guess not
Actually, this is better. You have to play with training rate as the model is already familiar with the concept. You most likely were overtuning your loras if you were trying to train fats. If you were training for style then it should have been fine.

>>4689
>Manual merge
Correction, this is supposed to be the manual score mode.

>>4687
>The novelai training was far and away the best, which I really didn't expect- you'd think that a lora trained using your model and used on your model would be the best, but I guess not
I had the same experience with it, glad I'm not alone.

Also, wow that enormous mumei ass...

>>4691
NTA I was trying to train a style LoRA, and NAI turned out better than, say, AOM3 or whatever BiggerGirls model I had also tried. Maybe I should revisiting training?


>>4689
Aurora? I'll have to check it out. So many models to try.

God, I don't want to be an asshole by recommending this while you're in the middle of your process, but I would recommend (in the even you were not aware!) using BGv3.5, which is a straight upgrade (upscaled training data, a lot more pruning). Possibly dream-irl or modelB (hadn't tried model B until very recently, and was impressed). That's also what the model maintainer recommends.

Regardless, I'm very excited to see the result of this merge.

>>4687
Janny being a faggot as usual. Catbox for that first Moom please? And the second too if it's not an inpaint edit.

>>4693
>using BGv3.5
Again, I have had better merging performance with my bbw-30k-step BG3 mix, which forms the basis for all of my experimental merges, and 3.1 Final. I think 30k produces significantly better looking fats (and on an anatomy level as well) than BG, and the variety problems that 30k has is solved by adding BG.
dream-irl is a photorealism model, which I don't want to touch given that could significant affect model artstyle.
My complaints about BG3 seemed to apply to 3.5 to a lesser extent as well. I just don't like the way it looks compared to 30k.

>>4693
>>4695 (me)
Look I'm not saying that the newer model might have better performance overall, I'm just kind of set in my conclusion that bbw-30k-step on average produces images that are better than BG, and is an overall easier model to work with in merges.

>>4694
The bullshit jannying is real. Very arbitrary.

>>4695
>>4697
If I understand correctly, you say you're adding BG to bbw-30k-step. In that case, wouldn't you be better off adding v3.5 rather than v3? I'm not suggesting to drop bbw-30k-step.

>>4694
alas, they're all inpainted
finals: https://files.catbox.moe/6nzwpr.png https://files.catbox.moe/m3k4jd.png https://files.catbox.moe/qvypv2.png
originals: https://files.catbox.moe/9kbr88.png https://files.catbox.moe/plgljk.png

>>4691
I might try again, then. I'm pretty satisfied with novelai though

>>4699
I'm more familiar with how this reacts when merged, so I wanted to go with something I know.

>>4701
Okay, cool! I just wanted to let you know that 3.5 was out (I was aware of dream, but not that 3.5 was released). I don't mean to say "you should use this".

Furthermore, with Aurora... how would you merge it will HLL? Standalone, nothing BBW, I mean.
1) Add Diff 1.0 HLL Aurora NAI ?
2) Add Diff 1.0 Aurora HLL NAI ?

something else?

>>4702
Aurora is supposed to already be able to do chuubas well. In the recipe they merged HoloKuki 2, at 25% which is another model merge but it contains a custom Hll derived from a merge of hll1 and hll3.1. I would be incredibly careful when trying to merge hll into it.

>>4703
Ah, I did not know it had chuubas in it. That's exactly the kind of information I was looking for. Thanks.

>>4702
>>4703 me
Also, rule of thumb, you should never merge models into hll, you should always be merging hll into a model.
(my recommended settings would be Aurora + 0.25*(HLL4 - NAI) if you want to try and increase chuuba presence.)

>>4700
> Original: https://files.catbox.moe/plgljk.png
> Final: https://files.catbox.moe/qvypv2.png
Holy fucking shit what a glow up. How? Is there a guide?

I could have tagged "SSBBW" this entire time and got results....

>>4687
So I'm a complete amateur when it comes to AI stuff, but I'm fascinated by the first two pics here. Which model/LORAs did you use for them?

>>4629
https://files.catbox.moe/qpv4ca.png

Forgive the mess of a prompt, I usually keep adding and forget to remove stuff.

>>4705
thanks for the advice!

>>4721
so wide... catboxes?

>>4720
I really like this size and shape but I don't care for how it blends with vtubers.

>>4722
https://files.catbox.moe/smgia2.png
https://files.catbox.moe/pug60b.png
https://files.catbox.moe/4ir923.png
https://files.catbox.moe/gtq0io.png

>>4600
Some preliminary shots, comment on size, shape, details you like to see etc and I'll do more

>>4721
The backgrounds look great! Did you do a lot of inpainting for those?

>>4736
Can you catbox these? And are these just with the hll one?

Why do so much fat artists and their subs hate AI? They ban AI’s users and their followers, write bad comments and opinions about us.

>>4741
Why do you think? They're mad because you don't need to go to them to commission your specific character as fat.

And where is this activity happening? Are you talking about DeviantArt?

>>4742 (me)
I should probably put gifs in the sequences thread next time. Was still in the old mentality of there being only one thread on /bbwdraw/

Also, I'd like to specify that I don't think AI has really affected the revenue model for fat fetish artists. 99.999% of viewers are "free riders" who don't pay for the art. Only a tiny minority pay via Patreon or commissions. Those types of people are going to continue to support the artists, since it's about more than just jerking off, which you can easily do for free.

>>4587
>your arts
I'm hesitant in calling my generations 'my art' because these systems are trained using the work of a plethora of artists I've followed for almost a decade now. As the outputs become more precise, the technology becomes more ubiquitous, and having to iterate or do anything in post becomes less and less of a necessity, the argument 'inpainting and photobashing still takes a certain skill!' is going to become weaker and weaker. These are going to become less like tools and more like content generators in a few years. Hell, that's what they practically are now. Any website currently hosting art has seen an explosion of generated images clogging up portals, from Pixiv to DeviantArt to R34, all drowning out the meager output by actual individuals and decreasing the value of art wholesale.

And it's not just art, it's been happening to writing and videos as well. A new side-hustle that cropped up on Amazon is generating novellas and children’s stories using ChatGPT in conjunction with Stable Diffusion models and selling them for five or ten bucks a pop. There was a minor trend of videos where the gimmick was ‘X as an X’ and those don’t seem to be stopping anytime soon so long as they garner a meager audience and create ad revenue. And among them all are books and videos talking about generating a passive income. Attempts to get rich with as little effort as possible. ‘How to Write A Profitable Book Using Artificial Intelligence with ChatGPT,’ ‘Content Creation Made Easy with ChatGPT,’ ‘ChatGPT Millionaire,’ and etc.

It’s been depressing me quite a lot lately. I think that the future of the internet is one where the opportunity that came from learning, pursuing, and sharing fulfilling hobbies is finally devoured by the hyper-disposable content swamp that’s rewarded for its sheer volume and output alone.

I’m trying my damndest to remain optimistic in lieu of all the new tech coming out. How they work is super fascinating but it also feels like a lot of what I hold dear about the internet may be on a death knell soon.

>>4746
Counterpoint (not necessarily less blackpilled than you, but in a different way):

On the subject of calling it "art", I agree with your sentiment, but I also think that ceding ground on this point (that AI images are not "art") is to weaken the position of AI images, including on legal grounds, which could lead to some very bad precedents.

At the moment, I would say that we're headed towards a new paradigm, where information and art can be mass-produced just like physical consumer goods. Did the industrial revolution hurt craftsmen? Yes, it did. Exceptional craftsmen survived, their work being prized because it was highly quality and not industrial, but the average craftsman was displaced by this.

Now, for the blackpills:

1) We could mitigate the impact of this disruption, but societies are too dysfunctional and corrupt to do so.

2) I mentioned legal precedents. This is where I think things could get really bad. A series of pro-corpo decisions could mean that generative tools are only in the hands of very narrow corporate or governmental groups, who keep those tools in a closed-source state to themselves. They would benefit immensely from these tools, and everyone else would be left out. This is a much worse fate than just economic disruption.

So my take: you can be pessimistic about the future, but you should also try to defend your use of these tools. If you don't, you'll lose them, but pandora's box has been opened, so *someone* is going to have access to them.

>>4749 (me)
I should at least post some content if I'm going to write a damn novella. My brain's not very awake today.

>>4746
Counterpoint: Model Merging is what makes AI image generation Art. Without model merging, we wouldn't have the current fats models be as good as they are. There is inherent black magic in those that do model merges, fine tunes, LoRA training, etc, such that it is incredibly hard to get started if you don't already get it, or haven't spent a shit ton of time trying to understand it. In the future, we may have people sell model merges that do incredible work, that are basically unreproducable because the tools required, and the method used hasn't been released to the public, requires so much time and effort testing and changing individual things, or requires a god recipe to produce, such that if you aren't using a base model created by merging 32 other models together with individual MBWs for each model, you will never be able to replicate the results that model will generate. And that isn't even getting into generating entirely new models from scratch ala NAI.

>>4746
I really hope, that all fat artists will go out. Why? Because they hate us, their subs, they make only what they want and not we want. So, all those arrogant artists as bamboo, bws, belt, presi and other idiots will response for their opinion about us

>>4742
> Are you talking about DeviantArt?
Not, discord’s and twitter of bamboo, bws and other artists. I really hope, that we will replace them

>>4750
How did you get that belly roll grab on Kson? Catbox please?

>>4750
Catbox?

>>4753
Bruh, that's not cool, they're the ones making AI possible. We can copy styles and gen new characters but without actual artists making art AI will get nowhere.

>>4687 catbox on the last one, even if it's inpainted? looks glorious

>>4759
/vtai/ refugees have no reason to recognize him but this is the resident schizo of bbwchan, don't respond to him. he is genuinely mentally ill and spends his entire day every day posting complete nonsense on every board as fast as he can before a mod notices him. both >>4754 and >>4753 will disappear soon.

>>4759
Diffusion models just regurgitate existing art reversing white noise. No one prompting owns the work. The same problem with tracing all over again. If commissioners and fans discourage tracing, personally don't see the threat of AI art. Prompting made me interested in art again and want to start sketching again.

Merge is complete. Here's some Hires fixed versions of the prompt used during generation at a higher step and a slightly different negative prompt (I got rid of (fat legs:1.1))
Link for the model will be forthcoming. I will be doing some more test gens to see if I will try and merge again. I have included some interesting information regarding the model composition for those that are interested in UNet discoveries. It should be noted, this was a pure weighted sum merge, not an Add difference. (If I redo this merge, I will most likely do an add difference with the fats model and Anything v4)

>>4753
>>4754
I don't see a reason to even be a part of their Discords, or follow their Twitters (or be on Twitter in general). Social media brings the absolute worst out of people. Right now, all the artists are terrified that the Machine will replace them (to be seen; as I said, it's not like their paypigs are disappearing. In fact all their simps are circling the wagons around this issue). Some artists have always been immature and childish.

However, you're not making things any better with the whole "I hope you're all destroyed by AI!" attitude. All that's likely to do is piss off any fence-sitters. Let the artists sperg out and be the assholes; you don't want to out-asshole them. Maybe even financially support the artists who aren't being huge pricks about it.

Anyway, I think that's it for me regarding this topic; I want to get back to tubby AI girls.

>>4757
Funnily enough, I didn't explicitly prompt for belly grab, although many BBW models / LoRA tend towards it unprompted because of the quantity of artwork in which it is present.

If you want to explicitly prompt it, "hands on own stomach" is the official Danbooru tag, but "hands on belly", "grabbing belly" probably help* (I haven't empirically verified that).

In terms of overall technique:
1) I use https://github.com/mcmonkeyprojects/sd-dynamic-thresholding to set my CFG very high (60!). In my tests, raising CFG causes better hand anatomy and poses.
2) The model probably helps. It's HoloCreamsicle (Add Diff 1.0 HLL4e3p8 AbyssOrangeMix3 NovelAI ) with 50% Bigger Girls Dream-irl mixed in on top.
3) I generate at a relatively high base resolution. 768 x 896
4) After comparing upscalers, I found the anime ESRGAN type upscalers (like 4x-AnimeSharp) to be best at not fucking up hands. Latent upscalers can easily turn okay hands into tumorous hand-gore.
5) For style, I tag for artists built into SD1.5 (e.g. Dean Cornwell), and combine that with generic anime artist LoRAs, and BBW artist LoRAs -> https://mega.nz/folder/sShj1LpS#Zh5ZxGO3uwK4QEaZ2X6z1w

>>4750
https://files.catbox.moe/lei2he.png
https://files.catbox.moe/j6vls8.png
https://files.catbox.moe/zey8rt.png
https://files.catbox.moe/hcw519.png
https://files.catbox.moe/0a757h.png

>>4763
>Diffusion models just regurgitate existing art reversing white noise
This is partially not true, and not fully how Stable Diffusion works.
While yes, diffusion models reverse white noise, they do not store art, nor do they understand art style. UNets only contain noise information, and the clip encoder just decides what UNets to light up when a token is passed through.

It doesn't inherent COPY any specific art style. You can train a model to copy and art style, but that is different.

>>4764
Model Link.
This isn't going onto the huggingface repo yet. So get it here. This should be compatible with Colab for retards.

https://pixeldrain.com/u/Y3s4CBg7

>>4766
Thank you for the box, info and loras link. Can I trouble you for a catbox of image 3 of this post as well please, the smaller very cute Moom belly?

>>4766
How did you add 4x-AnimeSharp to Automatic 1111? Can't find a folder to add it.

>>4769
no problem
https://files.catbox.moe/7n3olv.png

>>4770
You add it to:
...\stable-diffusion-webui\models\ESRGAN
and restart the webui program (not just the UI)

More temple series... I have a lot of these. I don't think I'm going to give them out in a huge zip, because that way I always have more material to post when I don't have anything else!

>>4766
> Maybe even financially support the artists who aren't being huge pricks about it.
All famous fat artists hate AI. So I can’ t do it. Does anyone have patreon or other pay method to pay for arts with AI. You should have your money
> any fence-sitters
They saw artists’s reaction and forgot about them. Just open YouTube and you can find, that no one uses artists’s work. So, artists just deleted themselves, hehe

>>4771
>I don't think I'm going to give them out in a huge zip
I am on my hands and knees begging for the zip.

>>4759
>We can copy styles
No, we can make new styles which inspired by others, as artists do this. Just see on style of bws and other artists, they look similar on all, that we can see.

>>4775
Well now I feel bad. Here, as a compromise you can get a 190MB zip of the series before I switched to using dynamic CFG thresholding.

https://litter.catbox.moe/xmpk1p.7z

Examples of that zip attached.

>>4766
Is there some config that you need to change in order to set your cfg that high? I can't get it to go higher than 30.

>>4768
Quick heads up, you may need to change your prompting style for this model. It won't automatically produce fats if you just put in fat:1.1 or something. You need to be specific in what you want to get what you are looking for out. So for example, the first picture is just the following prompt:
(extremely detailed CG unity 8k wallpaper, best quality, ultra-detailed, best illustration,SLE, best shadow, an extremely delicate and beautiful), ookami mio,orange eyes, BREAK black serafuku skirt, black serafuku top, black detatched sleeves, crop top, miniskirt, red choker, red neck tie, pleated short skirt, white thighhighs, bare shoulders, (wide hips, thick thighs:1.1), fat, in a room, detailed background

The second image is that prompt but with (fat:1.1) instead of fat.

The third image is that prompt but with (belly, fat:1.1) instead of just fat.

The forth image is that prompt but with (fat legs, wide hips, thick thighs:1.1), chubby, (belly:0.4), replacing (wide hips, thick thighs:1.1), fat,

I actually view this as an improvement since this means that you can get really specific and get correspondingly specific results.

>>4778
It's in ui-config.json in the root of the webui:
"txt2img/CFG Scale/maximum"

>>4779
Good level of control over fat proportions? Very well done, I salute your efforts.

Time for some really big chuubas.

>>4753
Everyone hates bbw-chan. I've seen this place shit talked whenever it's brought up on any other fat related forum or thread for years now.

>>4751
I do see how this is transformative. I guess where we differ here is how we use these AI systems now. I've been repeatedly finding myself in situations where I just turn to character.ai's chatbots whenever I'm horny because it's just so easy. As image generation becomes more precise and effortless, I could totally see myself someday just hastily slamming prompts together to generate content that'll satisfy my vices. I'm not saying it's impossible to be creative or transformative with image generation but I do believe it's going to be *really* easy in the future to practically let it run on autopilot and just indulge.

What I'm using these systems for specifically, I wouldn't call art, but how you're using them probably differs I should've realized that.

>>4783
>Everyone hates bbw-chan
The entire site is just beggers demanding free shit. I hated most threads too because I always paid for stuff and when I shared I only had people beg for more or complain that the shit they were getting for free wasn't good enough. No respect for artists here. AI is cool though, and I love getting feedback. Plus it doesn't cost anything to me.

This shit's way too nerdy to ever be considered cool. You all ought to be ashamed of yourselves. I'm going to watch a BBW orgy now and imagine myself as a transexual then jerk off all over my left leg and not wipe it off. You all make me sick.

>>4784
Yeah, I don't really like the rest of bbw-chan either. The AI threads have been really friendly, cooperative and comfy though.
Also nice hyper-thighs / ass.

I'm experimenting with gif2gif. Decent results, but increasing coherence in the background will be an area of investigation.

well if a zip of all the new stuff is out of the question then i will have to ask individually. requesting catboxes for:
>>4766
1 and 4
>>4771
3, 4 and 5
>>4777
1, 3, 4 and 6 please. all of these temple images are amazing

>>4777
Good LORD there is some gold in this zip.

>>4789
I was concerned that I'd run out material, but I have other content saved up for a rainy day. You did ask nicely twice...

https://files.catbox.moe/8jctt1.7z

>>4790
>Good gracious, ass is bodacious.

>>4791
thank you! i'm sure i am not the only one who is happy you decided to share and i will harvest this data for good use. nice shuba as well

>>4794 I'm sure and I so trust that you'll put ut to good use like only youcan. Thank you, endlessly for your efforts they're so appreciated.

>>4740
I used bbw_hll3 (a vtuber mix) and added the obesegirls lora I've been using. Seems to be a good combo for blobs
https://files.catbox.moe/rsuiuh.png
https://files.catbox.moe/973a7m.png

How do I stop the double-navel curse. Seems to be an upscale thing

>>4798
I've never found a consistent way, unfortunately, but it seems to happen much less on modern models at least. And I would much rather have that than the old stuff where it would turn them into nipples every time. Amazing size and shape on your first image, by the way, catbox please?

Wow it's true what everyone's saying, anime has never been this good or has never been better. A new pinnacle of human achievement has been reached here, guys.

>>4800
Yeah, I get it more on bigger girls, I think SD sees enough belly for two navels

>>4794
Thanks fren, that duck was also a favorite.

>>4798
Impressive size. My trick for navels is to put "navel" in the negatives. With emphasis. I also put "(multiple navels, multiple belly buttons:1.3)" in there, but honestly that might be cope. I need to validate it with a x-plot, but I've been busy.

>>4803 I could kill for some duck soup right now. I love duck soup.

>>4803
Yeah, this size is just bbw150g + lora, it isn't the best way to get big sizes but I like the more "compact" look of overwhelming fatness. Hard to explain

chubby moona

>>4784
>No respect for artists here
Based. They must shut up and forget respect

>>4798
Which model did you use?

>>4302
I think I’m starting to get the hang of tag and Lora weighting. I haven’t really read up on it I just mess with it as I go along.

The eigaka Lora is crazy good and plays well with Loras

>>4806
>bbw150g
where to find this model?

>>4815 >>4776 >>4774
cringey esl larp & obvious bait
>>4762 >>4762 >>4762 >>4762 >>4762 >>4762
> this is the resident schizo of bbwchan, don't respond to him. he is genuinely mentally ill and spends his entire day every day posting complete nonsense on every board as fast as he can before a mod notices him

>>4717 >>4760
The catboxes are here >>4700 , you can put catboxed images in pnginfo to find my settings and model >>2983 (Cross-thread)
The mysterydad lora is here https://mega.nz/folder/mUMxARJI#Wf5mGjsp6SchZArezJZJGQ
and the model is here >>4299

>>4706
there *is* a guide! >>7 (Cross-thread)
In essence you regenerate parts of the image to get what you want. If the model really has trouble you can help it along by drawing what you want or copying&pasting off of google or from another generation
It takes a lot of the rng out of getting good pics

>>4814
big mooner !

>>4826
wrong link, sorry
model is here https://huggingface.co/mariaWitch/ExperimentalBiggerMerges/tree/main

>>4822
Check MariaWitch's repo on HuggingFace, there is also the model madness thread here that contains a lot of models.

>>4826
What wrong i said about styles? We can make new styles, just use modern not fat models and obesegirl's lora for example. And also, you should compare those two pictures and say, which have better style and can be named more professional

I'm gunna fucking shoot up a wal-mart




In minecraft

>>4788
>The AI threads have been really friendly, cooperative and comfy though.
I think it has something to do with the fact that there is a not so insignificant number of us who are /vtai/ refugees. (Which is probably the best AI thread on 4chan in terms of quality, and sharing prompts, model merges, and general comfyness)

Could you send the catbox for these?

>>4832
Yeah, from what i've seen for now it's pregnant, cameltoe and on/off architecture that gets banned. With architecture its some unknown kilo you're not allowed to pass

>>4279
Could you catbox these?

>>4832
Did you get banned for having a belly???? This is beyond parody, lolwut

>>4832
this is retarded

>>4832
Shit, that sucks anon, thanks for the catbox on the yuge lioness, it was great practice for inpainting tails. ears are still difficult

>>4840
Looks good, I plan to do the same, I need inpaint practice.
https://files.catbox.moe/dsfzi2.png
https://files.catbox.moe/r50499.png
https://files.catbox.moe/79fa8r.png

>>4841
Good lioness butt, where did you get the style_mysterydad_v3 lora?

>>4622
more tests.

When it comes to huge hips and ass, GM&OH have a tendency to go a lot smaller than basil and GAME. Unsure if this is because it focuses much harder on composition and good backgrounds or if it just incapable of going much larger. But they are pretty good at backgrounds and composition

Basilmix is close to just being a direct upgrade to GAME and the best(imo) of the batch. Does all the same stuff while being better at backgrounds, poses, and details, while having a very "base" style that makes style loras play well with it. but theres a few things that are off:
1) the faces on really fat subjects are horrible (not a huge issue, actually, since it inpaints away fairly easily)
2) there's errors on very large sizes that GAME performs just fine at
3) there is a lot of noise / grain. it's especially present in highresfix and is more present the more fat tags you use

now to try the auroramix!

>>4840 >>4841
nice

>>4846

>>4840
Looks great, I need to practise that too>>4844
I think here? I'll look again later

>>4819
Catbox please? I love this.

>>4846
Hello anon, before you test the Aurora mix, make sure you are using the recommended Aurora Negative Embeddings as well. (They still provide a significant uplift in quality.) Also, make sure to use the default Aurora negatives.

Another thing to remember is that there are special tags in Aurora such as SLE and mksks style, which have a significant impact on the image.

mooner

>>4841
I'll try the other 2 as well. Just started with the hardest

Some big Silvervale and my second favorite squish, exercise ball

>>4856
thanks for letting me know, i'll run some XYs
just found the documentation https://huggingface.co/SweetLuna/Aurora

>>4861
catbox for 5?
>>4859
and 2 & 5
>>4857
& 1 & 2?

>>4863
https://files.catbox.moe/5sgaxh.png

>>4864
Could you catbox 2 and 5 of this post as well please? I love this belly shape.

>>4867
https://files.catbox.moe/zw8wtj.png
https://files.catbox.moe/ygmnxp.png

>>4841
and thats all. Was a learning experience

>>4868
Thank you!

>>4844
It's here >>4826

It's really fun for shortstack/giant thighs

>>4826
Fast question is this a stable diffusion extension or another web?
And also name of this