Also, if you feel like it, write captions for images posted here. We're all still learning how to use these new tools so don't fret about posting stuff that doesn't look quite right. Sometimes things look weird yet fappable. Prompts, help, suggestions, ideas — all welcome.

Sorry if some of you don't like the wet blueberry texture on these. I was experimenting with "detailed wet blueberry skin texture" and prompts like that. It's fun to think about whether these girls are getting bigger or if these pics are post-juicing.

They're a little more fun if you make up a narrative to go along with them. Example: In this one she looks like she's nearly done and trying to pull the juicing tube out.

In may cases, if you prompt for >body paint along with >detailed blueberry skin texture you will get something like these. Asking for a woman comes last after those prompts. Note that Stable Diffusion does understand the term >thicc but you must add a negative selector against "fat" or you'll just get obese chicks.

It's been very tough so far trying to get a full blueberry but for now I can get bits and pieces. Achieving a full blueberry will probably require img2img, but even then I don't know what to prompt for. These are more texture studies. Second one is a good example of an image that doesn't make actual sense but is satisfying in a certain way anyway.

More tiddies. I can try for smoother, drier texture if people prefer that. >just realized I should probably use tripcode for ID

I made a blueberry inflation as if it's from a R rated horror movie she's 24 year old

how do I find this program?

Love to see some more work in the blueberry girl/inflation style. I had trouble getting it to work in SD and Craiyon. Here's the best AI work I could do so far...Hopefully not the last.

The craiyon stuff is not as pretty but it does get you closer to the desired roundness of a blueberry girl

>>62258 https://huggingface.co/spaces/stabilityai/stable-diffusion

Variations of the OP >>62258 https://rentry.org/voldy You can run it on your own hardware if you have a sufficient Nvidia GPU.

More variations of the OP, some with a plain-colored face. I kind of liked the one with the one leg not really inflated; she's filling up asymmetrically.

>>62297 #78 is getting closer to the desired roundness. Has anyone been able to get full/mid transformation results yet? Around this reference size

I found this while creating a winter monster woman ,

It seems like any blueberry inflation created with Stable Diffusion is going to have to be done with img2img and never just text2img. The process goes like this: >find or generate original image, probably of a particularly thicc girl whose image will be easy to morph >use the liquefy tool to expand her >change her skin to blue >run that image through img2img using appropriate text prompt Specifying blue skin doesn't do much because SD doesn't get it. You just have to keep the denoising strength below 0.7 or so and it will leave her skin tone the same. Three variants attached to this post, plus the first image created by text2img.

Same process here. In this case, I just wanted a blueberry butt that seemed to be stretching its way out of a dress. Original at the end. It can help to use Waifu Diffusion for the original image, then Stable Diffusion and different artists to for the img2img to make create more details. Also, I changed the frame during img2img to be slightly wider. In this case it works to inflate her a bit more. Fun stuff.

>>62790 Wanted to add this one too. Her form isn't much different from the others but I enjoyed the fact that since I mentioned "blueberry" the AI decided to put a produce sticker on her leg. Now my fantasies about adult Violet include them slapping a big produce sticker on her, labeling her as a blueberry with a barcode and everything.

Image references are a must. Still trying to crack the code for consistent berry content

>>63858 Do this berry girl but Anna Taylor joy lol

How goes the experimentation? Anything new and exciting happen?

>>65394 I'm beginning to get somewhere.

Interpolation video tests. These are made by generating frames between one prompt and another, and then stringing them together into a video. They're never going to be smooth or any sort of replacement for an actual video, they're just fun and experimental.

One more. Was particularly pleased with this one. Again, not smooth and not a replacement for actual animation. Just experimental.

>>65971 Honestly I think you are pioneering how to exploit this new technology to this niche and I thank you beyond what you could imagine.

I've been playing around with it a little, granted this kind of inflation really isnt my fetish so been a little half assed

>>66002 Nice work, anon. I especially like #2 and #3. >>66004 Have you confused me for someone else? Also, I don't know what your image has to do with anything in this thread.

>>66008 Just memery Great to see what the AI can do

Id post what ive got but for some reason r9k keeps stopping my posts. Ive managed to actually get some decent stuff...

>>66842 If it's r9k then it's stuff that's already been posted. Would be helpful if r9k actually told you which image it was.

>>66874 True Punch/Wells (you posted each other’s work you’re the same person lol) 🌕

OMG, how can i make my own?

Bump!

Dumping some stable diffusion img2img stuff I generated

>>69285 There appears to be several different 'walls' we'll need to get over somehow. the first being body shape. The second being skin color. The third being clothing. I break these apart because the models are going to struggle with each. It's why blue skin is probably easiest to train, whereas clothing will be hardest. And it's why body shape seems to default to massive obesity rather than a traditional shape. But as other people have said, references are going to be the important bit.

Was testing to see if the 'everythingv3' model could do anything, seeing as it's trained on a lot of anime stuff. the blue skin tag seems to make them almost translucent? More water balloon than blueberry. I'll play with it more, see what else can be done.

could anyone do a blueberry expansion chiaki from danganronpa?

So I have been alerted to this, and while I cannot use it yet myself because I lack a needed gpu with between 9-12 gb of vram, this may help a lot of other people. https://github.com/7eu7d7/DreamArtist-stable-diffusion It also has an automatic repo extension that you can download it from. Supposedly, it can create an accurate embedding with a SINGLE image that can be used for various new creations. I say supposedly, because I personally can't use it yet, and I don't see a collab yet for it. But, if it works as advertised, perhaps this would create real progress toward the ultimate goal?

Got some more good results with img2img and careful prompting

>>69634 I've found that img2img can produce some great results, once it has a guideline for what It needs to produce. I never tried doing blueberry stuff with it in terms of img2img though. I'll need to give that a try.

>>69636 Mate, using ai modeling isn't 'abuse' any more than using photoshop instead of a pencil is 'abuse.' Times change. Equipment changes. And given how diffusion works as a process, you're only ever going to replace the bottom 10-20% of artists. Diffusion isn't abuse anymore than people using terrible photoshop to put people's heads on movie stills of blueberry violet is abuse. Tone it down, chill out, accept the future as it comes. Can't be fun being mad all the time when there's fun to be had.

>>66320 >>69659 amusing notion, given no ship is sinking. The future will come, no matter what reactionaries believe. You can no more stop ai being added to art anymore than you can stop the moving to 5g. leave your fearmongering, reactionary tripe elsewhere. There's already a thread for manbaby whining.

Recent edition from my works:

I found this site just now; it’s slow and too busy most of the time but the results when it does work are pretty good. https://h5.tu.qq.com/web/ai-2d/cartoon/index?parent_trace_id=7f6d9f4c-3301-6461-b8a4-0f5ab7a5668b&root_channel=&current_channel=imageQRCode&level=1

At the moment, I'm trying to experiment with hypernetworks and textual inversion embeddings involving blueberry content; results are mixed right now. But if I get good results, I'll post them here. I suspect that in order to get good blueberry results, a lot of training is going to be needed.

So, this is an unfinished training image. It's got a ton of flaws and is very much a wip. But, it's what I've been managing to do while training the textual inversion on various blueberry images. The process is obviously slow. But the fact that it's possible to get it to understand the correct shape, sorta, is progress on figuring out how to make this sort of thing possible. I'm hoping that I can find a way to mix this training with other things to get good results as output. Also, perhaps other people will take this lead and be able to generate their own better things, because I admit I'm often behind the eight ball when I'm generating content. But it IS possible through textual inversion, at least. It may also be possible through hypernetworks.

Still not much luck for me with it comes to full body inflation. I occasionally get lucky with balloons but blueberries are quite difficult. Maybe this belongs more under Blueberry Faces but I thought this is closer. These are, as you can see, based on the musicals. Photoshop was involved so it's not just SD. I tried to capture that moment when she's starting to feel a bit weird and notices the Oompa Loompas are starting up with another song…

>>70380 This is awesome. Where can we find the rest of your stuff?

>>70866 Thanks, you do great work on DA. You can come say hi over here: Twitter: @CurvyDeluxeAI or @stuffeddeluxe DA: stuffeddeluxe

Still tinkering with Blueberry stuff. Have to lean into the img2img references for better results.

>>66320 >>70865 I'm certainly liking what you're doing in terms of outputs. I think you could probably tinker with it a bit and get what you want? But it's true, the further away you are from 'the norm' the harder it is to train. I have not, for example, been able to get anything close to that color of skin; I am curious what methods you used to train it as well, especially whether you're using dreambooth or not.

>>71863 I do nearly all of the blue skin in Photoshop afterward. I really need to remember to include that in descriptions.

>>71994 >generates a cute girl with a phat ass aweso- >photoshops her skin blue

>>72133 >something was not made specifically to please me I have the non-blue originals which I can post in the SD thread on the next drop though

>>66320 Been experimenting a bit with img2img and blue skin; in general, I find that most of the generators are absolutely terrible at generating any kind of blue skin. However, if you color it in photoshop and do a few img2imgs, the models seem to grasp what you want easily enough. Going to have to test it with other models though, since I've only managed it with AnythingV3 at the moment.

bump

made this with biggergirlsmodel, edited to get blue skin

tried to get more realistic berries

>>72546 Huh, seems promising. Are those images examples of the photoshopped ones you're feeding them or stuff they spat out after?

>>66320 >>74530 Not a huge amount of process to these, as they were tests. Basically, I generated an image, did the normal blue photoshop overly, and then used a really low denoising img2img to basically get the same thing out. I think at the moment, the blueberry faces thread has a bit more up to date example of what's done, since this was done a while ago now.

Zero AI experience but a lot of photoshop/morphing experience. Can I use AI to generate boobs and bellies to use in my morphs? I'm vaguely aware there's something called 'inpainting' - could I use that to 'paint' bigger bellies/breasts in to images or am I understanding it wrong? Basically want to use AI and photoshop to create that spicy good shit but have no idea where to even start with the AI side of things.

>>74905 This is the best guide for getting set up: >https://rentry.org/voldy You could definitely experiment with using SD to generate something to work from. I think it might actually work best the other way around, creating the basis image first and then using img2img or inpainting to generate nicer details. Inpainting is when you select just part of the image to regenerate. You can set the AI to mimic the original to a certain degree. I use this for faces a lot. Tons of tutorials on YouTube, that's where you'd get better information. But really you just have to mess with it, same way you learned Photoshop.

I've playing with models recently, got some different results that turned out alright

Just testing it.

>>62189 One AI art thread is more than enough.

Getting fairly good results here with Deliberate v1.1. 🔗 https://huggingface.co/XpucT/Deliberate Use terms like (((dark blue blueberry skin))) with a lot of emphasis, and at the beginning of the prompt.

>>77325 Dude, these are fucking incredible. I will absolutely be trying this.

>>78014 Noice

Some fresh berries. >>77509 Thank you 💙 >>78014 Wonderful, I love these 💙

>>78014 I am curious what model you're using, what prompt you're using, and whether or not you've been doing any training of embeddings or models and the like to get those shapes.

>>78576 Just using NovelAI, not a custom model. Specifically, using the NovelAI furry model, but forcing it into humans tends to result in a less anime style and more interesting posing in my experience.

>>78612 As for the prompts, I unfortunately didn't save it at the time, but here's my attempted reconstruction of it: Prompt: {{{human, female, purple body, monotone body}}}, {{{nude, 5 fingers, brown hair, long hair, holding own belly, {{big belly, hyper belly inflation, big thighs, big hips, aroused}}, breasts, solo, blush, seductive, big eyes}}}, {inside, kitchen, warcraft}, {{digital painting (artwork)}} Negative Prompt: pregnant, muscular, vore, signature, male, {{elf, anthro, noseless}}

>>78617 Damn can you do celebs like that?

Been doing a series of blueberry girls getting juiced or coming out of the juicer. They never get quite back to normal, but that's what you get for chewing gum all day long. Sorry about the watermarks on some of them but it turns out some people are reposting without attribution. I don't care most of the time but sometimes I spend enough time on a pic… BTW feel free to repost anywhere, just credit hgswells9000 and that's enough for me.

>>78969 Oops, one more (the one that took the most time). Poor girl got hoses in the front and the back 😰

>>78969 holy... your generations have come a huge way ever since you started working with them. Absolutely amazing work here. I would love to know your prompts. Your img2img and inpainting is absolutely astounding. Truly great work.

>>71984 >>78969 Also, would you be willing to post a more detailed tutorial for this, with example prompts, on here or dA? I have the two you've made before, but I've never managed to get anything close to this with the method, and doing various experiments with the x/y plot with tags like (dark blue blueberry skin) at various strengths only seems to make it want to prompt lots of actual blueberries. Too little strength, it doesn't appear, too high and it seems to want to generate plants. And I am using the traditional blueberry color photoshop method and img2img. It's just not coming out appropriately. I'm also using deliberate's newest model, as you are. So I'm certain I'm probably messing up either the generation method or the prompt itself. I know I wouldn't be the only one who is genuinely curious about what prompts you use and how you generate these things, because they're head and shoulders above what else I've seen.

>>78612 Where do you put the furry(1,2,3).pt files? I cant seem to find their desired place

Can't help there - I was using the actual website, not the leaked models locally.

>>79167 are you talking about the hypernetworks? If so, they should go in the hypernetwork folder.

>>78969 oh, one last thing. If you saved your images as PNGs, then we could load them into the webui and see the prompt you used, so we wouldn't need to ask.

Best results I've managed thus far. Not much, but progress?

deliberate

Continuing the Juicing Room theme as it seems to be popular. The inpainted hoses here look a little out of place but I left them in just to get feedback. Plus, you know, the implication… >>79234 These are great, especially the middle one. 💙 >>79437 Well done, anon!

>>79553 I gotta say, this "series" of yours is fantastic.

>>79553 I seriously wish you would save your images as pngs so we could see the exact prompts you use. I hate asking over and over, but I'm genuinely curious about what prompts you use for the base before you start img2img-ing and doing inpainting. The reason is that I find it very hard to get the texture right. I don't get anything that looks nearly as fruit-like as yours.

Here's my go at it, I'm pretty new to stable diffusion, not nearly as good as HG Swells

Tried making some bbws and then editing them to be blue

bump

>>83380 This has to be the best one here holy

>>83384 thanks! i started with a pic i made with biggergirlsmodel and then used img2img with realdosmix to try to get it to look more realistic. i'm still struggling with img2img but i liked this result

been trying for a couple of days to get something half decent blueberry related and this is what i got so far

>>62189 >>>/bbwai/

MODS, please move this thread to /bbwai/, please and thank you.