Here's more generated using stable diffusion, with a realistic art style.

These looksblike SFM render

This is cool. How does it fare with generating non-human and non-anthro pregnancy? Would you experiment with Rarity a little more for me? Someone else claimed stable diffusion doesn't recognize pregnancy on non-humanoid bodies, but was able to get it to generate a big belly that was close for his purposes. You can see that here: >>1398

Here's another set I did with officer Judy Hopps. It can recognize outfits pretty well if you can come up with a specific enough/good enough prompt. This is really very good for generating a series of images of the same character in the same style - you can pretty much get anything from it once you get a nice prompt. It can also recognize and differentiate between hyper pregnancy, normal pregnancy, cumflation, vore, etc. Model used is Zack3D_Kinky-v1. Prompt (varies over the image set, but was used to generate the last, biggest pregnant image): judy_hopps, grey fur, 3d, realistic, ray-traced, big boobs, thick fur, thick_thighs, solo, solo_focus, navel, leather bondage_gear, bondage, bedroom_eyes, inviting, cute, pregnant, (big_belly, hyper_pregnancy), sexy

>>1464 I can try to run a few imaged with the "feral" tag. This model was trained on some of the kinkier shit on e621, so it should do well, but we'll see.

There's a lot of times when the generator makes a little abnormality, like the navels on Judy in hopps7.png. I should clarify, these are things that could be removed in the stable diffusion inpainting feature, I just don't really have the time to spend refining each image that much.

>>1464 Here's some non-human, non-anthro rarities for you. I had varying luck, but I wouldn't say it's overall any worse at building these pregnancies than it is at building anthropomorphic ones. It's just a little harder to spot a bad result on an anthropomorphic character - if the belly is a little misshaped or out of place, it's easier to notice on a feral model. I used "female feral rarity_(mlp), equid, equine, my_little_pony, all_fours, ((pregnant, big_belly)), realistic, 3d, ray-traced," as the prompt and "pussy, presenting, duplicate limbs, boobs, nipples" as the negative prompt, mostly because boobs gave it trouble on a model like this.

If anyone else has prompt/character ideas, please do post. I'll take a look and run a few.

There are models trained on pony content from derpibooru, etc as well. I haven't tried any yet, but I bet they'd work well.

Is there a way that others can try it? Or something similar to it?

>>1483 Yep! Anyone with more than 4gb of vram can run stable diffusion! Just find a tutorial for the latest edition of stable diffusion, follow it, and you'll have a version with the default model. You can then look for a model that fits your desires. There's a few furry ones, and a pony one, and a few anime ones. Download, and put in the models directory. It's free, open source, and relatively easy! All the hard work has been done by other people. You can even run it on a Google Colab if you don't have the resources for it on your own PC.

OP here. Figured I would walk people through the steps I took to start generating stuff. First step: downloading stable diffusion Before downloading, make sure you meet the system requirements. Type "dxdiag" in the windows search bar, it will bring up an information window on your system. Make sure you have at least 4gb of vram (integrated graphics won't work). Here's a link to the tutorial I used: hxxps://www.youtube.com/watch?v=vg8-NSbaWZI I replaced the "https" to "hxxps" so we don't flood the poor youtuber with visits from pregchan.com in his analytics. This tutorial covers how to use, learn, and navigate the stable diffusion webui, so I won't cover it here. Second Step: Installing additional models During the process of downloading and installing stable diffusion, you should have a folder named "models" in the stable diffusion webui directory. This is where the default model goes, and this will be where you put your additional models. You can put in as many or as few as you like, as there is an option in the webui config to choose the model you base your generations on. Finding the models was the hardest part (for me, at least) but it looks like this website has a big fat list of them: hxxps://rentry.co/sdmodels#zack3d_kinky-v1ckpt-1a75d5c6 Each model is quite large, upwards of 2gb, so be prepared to wait a while to download. My favorite is Zack3D_Kinky, it's what I've been using to generate the images posted here. Try a few, see which one turns out the best results for you! Customization is the name of the game here. Third Step: Generating Images You will find that the stable diffusion webui is a bit intimidating at first, but there's a few key things to keep in mind. 1 - Set the step count to around 50-70. I find this generates the best results - too low, and it will be blurry and unrefined, too high, and it will take long and give the same results. 2 - Pick detailed prompts. Start general if you must, but the more specific and detailed your prompts get, the better the output will be. I often start with something like "applejack, equid, equine, anthro" and see where it takes me, eventually ending up with a much more complex prompt. 3 - Use txt2img to generate batches of low-resolution images. This lets you cherrypick the best results and take them to img2img for refinement and upscaling. 4 - When using img2img, you can either create a batch of several images all based off one image, with high "denoising strength" to create a variety of images based loosely off the original prompt, or you can create 1-2 high resolution images at a low "denoising strength" for simple upscaling. 5 - Toy around with it! Flick switches and slide sliders. You may find that something changes in an important way with a slight tweak to a slider. You can also use prompt matrices, styles/artists, upscaling, etc. More information here: hxxps://github.com/AUTOMATIC1111/stable-diffusion-webui-feature-showcase Good luck out there. I'm hoping with more eyes on it, stable diffusion can be refined even further. P.S. - although I haven't tried it yet, I'm certain the inpainting feature could be used to edit images into preg variants. Someone else can give it a shot.

Figured out how to upscale to 4k. Takes a while.

I'm getting a lot faster and a lot better at getting good results here. Again, if anyone else has prompts/characters they'd like to run, post them! Here's some vaporeons

4k res is too much to upload all of them at once

>>1490 Finally got this figured out and trying some attempts with Lucario/Krystal, SSD seems better at properly identifying a Pokemon over an established character. This was the least cursed creation after about 10 batches of 8, maybe you might have better outputs.

>>1473 >Here's some non-human, non-anthro rarities for you. Thanks. I really appreciate it. If it weren't for the fifth leg in the first, that would be a perfect image of her, but all but the second are pretty good. It even got her mascara right. >I had varying luck, but I wouldn't say it's overall any worse at building these pregnancies than it is at building anthropomorphic ones. In that case, would you mind trying again without big_belly and maybe the horse tag if one would be applicable? Perhaps a run just without big_belly would be best.

say.. what kind of AI generator is? perhaps some website?

>>1507 I wanna know too!

>>1497 Here's one of my outputs

Bonus lucario, found this one while screwing around and oh my, it's flawless

>1507 >1508 Check out my post above, >1485 It's a local program, running on my graphics card. Anyone else can download and use the same thing, for free. Bonus lineart fox I thought was good.

>>1500 here's a quick one, keep in mind big_belly is a descriptor - if you want it to look pregnant, you have to specify it. This does have the pregnant tag included in the prompt.

>>1512 Oh, I see. Well, thanks again. I guess it's just a matter of rolling the dice enough then.

>>1513 Yep. Persistence is key. When you get a result that's close to what you want, send it to the img2img tab and play around with the noise slider there. I often use that to fix less-than-perfect results.

>>1497 Here's what I got after toying around with your image in img2img for a while.

>>1515 I think this showcases pretty well how this software can be used to generate edits. Take an input image, write a reasonably well-fitting prompt, and run it until you get something close to what you wanted.

Might i make a recommendation of this character and or others by the same artist love to see how they look in this AI? thank you very much. ^^

>>1518 Note sources are on e621 for the previous images, and also: https://e621.net/posts/2743546?q=lucy_%28hladilnik%29 https://e621.net/posts/2553542?q=shay_%28hladilnik%29 If anyone can give these 4 a shot, i would be very grateful for the opportunity.

>>1520 All 4, pregnancy to be specific if you be so kind? Thank you

>>1520 I won't be doing all four. It took about an hour to get inpainting nailed down, and another hour on top of that to get a good prompt going. Notice how the resultant images pretty closely follow the contour of the original image - which means that you could draw a crappy ms-paint belly on a character and use AI to make it consistent with the rest of the art style. Lucy's a good character, so I'll do another with another pic of lucy from e621 I would encourage you to download stable diffusion and try to pull off a couple yourself. Try drawing a belly on in ms-paint (even if it's REALLY shitty it will still help the program get the contours!)

>>1576 You're the man my dude for doing the Lucy, hope to see Shay and the others in time. Also cannot do stable diffusion, got a Mac so options are limited.

>>1577 Actually, you probably can. There's a version available via google colab that doesn't run locally. As far as using a furry model with that? It may be possible, I don't really know.

>>1520 Here's 2 more of lucy, plus the prompt image I used. 5 minutes in ms-paint

>>1580 don't wanna toot my own horn too much, but I think that last lucy is really good. the power of this for edits...

>>1473 > hxxps://rentry.co/sdmodels#zack3d_kinky-v1ckpt-1a75d5c6 That second one is just about perfect.

>>1579 Mind hitting me up with a link op? would really love to try it myself to get it running. :D

>>1485 >>1583 See 1485 for all you need to set it up

>>1587 Ok tried, and it did not work.

>>1589 Here are some things I learned the hard way. * If you download any models you have to reload the stable diffusion UI to access it. * Do not close the command prompt, this breaks the UI and you'll have to reload it, which means another IP to access it. Hard to pinpoint your problem with just "did not work" but this may help.

>>1589 Do you have an Nvidia graphics card with >4gb of vram?

>>1592 Ah shit, i do not. Fuck.

>>1593 Look into running stable diffusion on the cloud. It's possible, and it can be free on Google Colab. Not sure about using furry models with it over the cloud.

>>1473 Which pony model are you using? I've found a "pony_sfw_80k" and so on, but it doesn't seem very consistent.

Managed to get these two, out of nearly 100 abominations

hxxps://nmkd.itch.io/t2i-gui Here's a link to another Stable Diffusion client. You don't have to donate to download the client, but there is like 10 GB worth of stuff to download, both the client itself and supporting files once you have it installed.

https://pregchan.com/.media/8864d4444dd30dba9fc9e70f8d2e6150d1a36c28c8d1d3c5b4ae82450242a2a2.png Mayby someone could try with this tigress please?

>>1596 I have been using Zach's kinky 3d

Here's some

>>1635 Try a dolphin or beluga next

Could this be edited to be preg? I'm curious .

Anyone running a colab with this model?

had a hell of a time trying to get it to recognize what a pregnant lucario on all fours would look like

Mayby those two jackal girls can be edited?

Mafia

Dinosaur civilization

>>1657 Thanks, arsehole. Now I'm a scalie.

>>1657 Impressive results. Mind sharing some of your prompts/ the model you're using?

Would love if people posted their prompts with their pics.

dinosaur dinosaur dinosaur

dinodinodino

Here's a big repository of models to use with stable diffusion https://rentry.org/sdmodels

>>1655 first one is in a pretty unfortunate position, AI tends to do best when the subject is upright, because that's the most natural/common pose. If you felt like giving me an ms-paint sketch (color preferred, black and white works too) of what you want the belly to look like on the second one, I'll give it a shot. Try to outline the rough size/shape you want and color in the air pipe.

>>1657 what are you using to get that redness on the last pic?

>>1666 Just to be sure, it can be whatever pic with the wanted size of belly? I ask because im unable to use anything since i don't have pc, heh..as for air pimp, mayby it could stay like it is?

>>1690 Yeah, any pic. Drawing something in with MS paint just makes it easier to get a good result with stable diffusion. A rough sketch will do. Air pump can maybe stay, I'll try it with and without.

>>1704 https://www.furaffinity.net/view/17233802/ I found suitable sketch with the size i want this. Or perhaps this would be goo? https://www.furaffinity.net/view/2630313/

>>1637 It takes some work, but yes.

>>1705 >>1580 You're gonna need to edit the image you want edited. See >>1580 for an example

>>1706 Holy crap that is amazing! Whoever did it thanks a ton

>>1709 Unfortunately that is out of question, i don't owe pc so won't be able to edit myself :/ But mayby the belly on third pic would do? If not then that's unfortunate ._.

Any chance this can be improved upon?

>>1750 I couldn't get the styles to match. That's one problem with SD, it kind of does it's own thing unless you're willing to put in a lot of time to train it. However, doing it's own thing I got this.

>>1754 thats really beautiful! thank you so much!

>>1759 Middle Eastern residents of Zootopia

I really want to see a glamorous poodle lady with a belly that is ready to pop while wearing a skin tight dress. if anyone decides to have a go, I will be eternally grateful

>>1761 How about this?

>>1762 You have just made my dreams come true, thank you so much for this!

>>1762 Name of the software?

>>1764 novelai tag: Highly detailed，Render unreal engine，8k Resolution,pregnant,big belly, highlights (coloring),digital painting,gorgeous pregnant, sexy, big pregnant belly, tight belly,thick coating,glamorous poodle lady，skin tight dress,evening dress,intricate ornate details,white dress,round belly,navel piercing

>>1764 Do you know how to use NovelAI without spending money?

>>1770 You don't

Can we have some big bellied cocker spaniels working in a office? that would be fantastic

I'm curious how well the AI is at creating birth images, is it possible to have any of the ladies in >>1762 at home delivering their pups?

>>1774 Realistic style and zootopia style

>>1775 These are absolutely incredible! This AI is crazy! Thanks a bunch

Holy crap this AI is insane! It looks so good. I wanna request (to anyone that’s interested who knows how to work this program that) to try and put a big ol’ pregnant belly on Mewtwo, Hornet from Hollow Knight, a female enderman, and that blue fur alien from that new Kirby game named Elfilis i think. I know I’m asking a lot fro, random people online, but I’m really curious to see if it’ll work.

>>1779 Oooooh! That looks really good!

How about some full term country girls that are Afghan hounds

novelai is good

>>1796 These are great, if ifs not to much can i request one more thing, some lionesses and tigresses 9 months in sexy tight yoga gear, thank you.

Can anyone pretty please make something from this? Just add a nice belly, many thanks lads

>>1799 holy crap these are amazing! Thank you so much! You’re really good at this!

Goddamn this A.I is impressive! Can it magic up a elegant wolf queen in her 3rd trimester in a beautiful tight gown, very much appreciated if anyone decides to have a go.

>>1519 >>1518 If anyone can muster up some pregnant images with these characters that be awesome.

>>1815 more dragons

>>1816 and more dragons

>>1817 and even more dragons

>>1813 Oh yes she is perfect! Thank you so much!

>>1801 anyone, please, my SD gives out literal bullshit instead of these arts, even with the furry model in place

>>1820 Have you tried using inpainting? Have you tried sketching on a belly and then using inpainting? I promise, stable diffusion is more than capable of doing what you want there.

Can we get some big beautiful horse ladies rocking in some lingerie

>>1823 it gives the art weird glow for some reason

>>1823 Inpainting, super SD, using an already pregnant image as a base, all my outputs are visibly worse than the original. Legitimately don't know how you guys do it.

It's likely dependent on the model but how does SD fare with hyper pregnancy and/or navel focused stuff?

>>1826 Are you using the base stable diffusion model? Or one of the furry models? Have you tried more than one model?

>>1828 Using the super stable diffusion linked in >>1485, using furry_epoch4 and yiffy_e18 at the same time as my models.

I was messing around with novelai myself and I got this image of a zoroark. I have a few others but they probably aren't as good as this one.

So I just let it have its way making pregnant wolves, and got these three... I'm scared to let it re-process them with what I've been getting, so I'm just... going to pass these three off as nice results. Hopefully we either still have a thread, or another iteration of this one in about 6 months. I'm excited to see where the models and program go by next spring/summer, considering the progress this year alone.

>>1837 Try adding a few "quality tags" This includes putting stuff like "low res, normal res, jpeg artifacts, signature, watermark" etc into the "do not want" (or whatever it's actually called I forget) box, and putting "high res, absurd res" etc into the prompt. You'll get better results that way. You can also specify bad anatomy, bad mouth, etc in the negative prompts. You can also specify art styles. I find it does better with realistic styles than cartoon or lineart.

>>1829 Care to share a few of your bad results? I haven't seen someone use two checkpoints at once. I should also ask if you're using an appropriate prompt in inpainting, such as "pregnant, female, equid, equine, anthro, brown fur, fur," etc. I think we'll all be better off if you can learn to do it well instead of just a few people doing all the requests :). I also think you are capable of doing it well.

>>1837 Don't be scared to reprocess them. If you're using stable diffusion or novelAI you can adjust the degree to which the AI changes the image in img2img with the noise slider (lower = less adjustment). You can also use the inpainting feature to change specific parts without touching the rest of the art, and you can still adjust the level to which it changes the image there.

>>1830 here's my shot at it

>>1824 >>1827 two for one! hope that satisfies the itch. I was certainly happy with the results.

>>1842 These are great! This A.I has so much potential, big thanks whoever did this.

This stuff looks awesome! Is there a chance for scp 682, 1471? Or perhaps exveemon and raidramon? There's not that much preg art of ekans or samurott. I would be happy to see some if someone would give a shot. ;3

>>1839 First image is the base, was trying to use inpainting to make her even bigger. Unfortunately as you can see by the outputs the inpainting on SSD couldn't tell what to actually color the masked areas with so it just does whatever.

>>1846 I did the Lady edit (I also did the original Lucario edit too). The work flow for SD is to open the source image in a paint program and use the cream color of her stomach and draw out where you want the belly to be. It can just be a flat color. Load that into inpaint and mask the entire belly. If you just mask the area without drawing what you want SD thinks you want to redraw the background. The hard part is getting a prompt so SD can match the art style. It worked with Lady because that was a 3D render so there isn’t an art style (like line thickness or unique shading style).

>>1775 Can we get more of the realistic spaniels?

>>1847 Now that I better understand how masking/inpainting works progress has been made but it's still hard to make it blend in enough with the original. And the occasional output where the AI thinks the belly is a second set of tits instead like image 2. Could just be a limitation of the art style and should try with other images. Unrelated, your contributions to the pregedit threads is nothing short of exemplary, I think I speak for most of us that edit commissions are still on the table for you to consider.

>>1856 Check to see if the original image is in e621. If it is, look for tags describing the artwork or qualities of it and add them to your prompt. That may help.

>>1845 I'm using novelAI for this, it's actually possible that the furry models for stable diffusion would have more luck with the more obscure stuff. However, NAI did "recognize" scp-1471 and gave me a lot of results that looked like this.

>>1845 It's just tough to get a good result from tags that have less than 1000 hits on e621. Especially because of those 1000 images, very few are actually fit to train an AI on. I'm looking forward to a simplification of the training process such that it doesn't take 24gb of vram. When that happens, we'll see a lot more variation, and will be able to train it to recognize specific tastes much better.

Here's a few initial takes from novelAI. I'm pretty impressed so far, it seems to do much better than plain old stable diffusion as far as instant results go. Stable diffusion is still a better tool for inpainting, outpainting, etc.

Can anyone make a pregnant SCP-939? With and without breasts

>>1862 These are fantastic, Thank you so much

I would love to see a hyper pregnant/busty Krystal either in her Adventures tribal clothes or in a shrine maiden outfit if it's possible.

Since it's nearly Halloween, how about some witchy black cats for the occasion?

>>1867 They all look good even if the AI isn't the best at knowing how Krystal looks, thanks.

>>1873 oops i meant to post this here, sorry

>>1874 Yeah okay. Got a few nice results here, impressed with NAIs ability to generate bellies and hands.

>>1875 These are really good, thanks for doing these!

Obligatory dragons

I would like to see preggy female Spyro the dragon, cheers.

>>1878 I wonder how well it handles dolphins

I have two more requests. For this one I wanna see if the AI can do hyper pregnancies. Can someone try and do a hyper pregnant female Angel Dust or Hornet from Hollow Knight again. The 2nd thing I wanna request is to see if the AI can do a Mpreg Stolas from Helluva Boss and/or an Mpreg Grimm from Hollow Knight.

>>1886 Reasonably well, though it seems to want them to have blue skin by default. Had to specify grey body. Also seems to like giving them hair. That's not so bad.

Can the A.I handle some huge kangaroos in long dresses?

I'm curious, is it possible to make a pregnant edit of her with one of the ais?

>>1890 Definitely! Stable diffusion is free and works great for stuff like that. Give it a shot! I think >>1485 has a tutorial.

I have quite a bit to share

Almost forgot this one

How good do Espeons work on this?

>>1875 >Trunk arm on the third one Sometimes this thing sneaks in those weird little details

>>1762 It's me again, is there any chance we can see even more beautiful poodle ladies in long gowns or maybe office uniforms? Honestly if anyone did any of the two you will have my gratitude.

>>1909 Tried to get a comic out of it, got this. Also a couple good ones, more on the way.

>>1915 >>1914 All of these are absolutely fantastic! From the bottom of my heart, thank you so much!

Mayby this would be good sample?

>>1915 >>1762 Not the original requester but i really wanna see these poodles full term as teachers, they are smart dogs after all.

>>1920 To tell you the truth i would love to see that, they can wear buttoned shirts with skirts or trousers, that's if the A.I madlads want to do it, but if not then fair enough

>>1920 >>1921 Why the hell not. Second one is infuriatingly close to perfect. Others are pretty good. Enjoy!

>>1922 I have a question- how exactly does one end up getting into these? How do I go about training my own AI like this to produce results like this, and what steps need to be followed?

>>1922 It might not be my idea in the first place but i have to say, these look absolutely phenomenal! whoever took the time to do this, a million thank yous.

>>1922 Wow they are good, thx

>>1923 Check >>1485 For a guide. No need to train anything yourself. Both stable diffusion (free, and pretty good) and novelai (paid, and better) have models available for furry stuff.

>>1518 Oh yell yea, zig zag needs a good belly bump, maybe even birthing; anyone wanna try?

We need some heavily pregnant lion queens

I am curious, could this a.i magic up some husky brides?

>>1948 How slutty do you want?

>>1949 Very

>>1935 I can assure you that no AI yet developed can recognize and output birthing stuff. That's probably for the best, for now.

>>1952 Thanks for this man, these are great!

>>1952 >>1948 Even just some nice wholesome huskies would be perfect if anyone fancies a go

How well does it handle insects? I’d assume it’s not as good as the other stuff.

>>1951 Very well, hope to see some pregnant zig zag under the AI though.

>>1813 Would it not be to much trouble to ask for more wolves like this one here, i am a sucker for preggy wolves.

>>1948 I really want to see this idea happen, it has a lot of potential to be cute and sexy.

>>1706 Question. Could these be edited like the Lady pic?

>>1948 A few results

>>1982 Thank you! These are some sexy huskies!

>>1982 lol @ the 3 arm'd husky

We have seen poodles that are fancy and some that are teachers, i want to keep the poodle train running, for the lols how about some wearing swimsuits?

>>1986 >>1982 Please try using negative prompts. No reason to have mutilated watermarks in your generated images. Would be nice to have a few others actually contributing art in this thread. It takes very little actual technical skill, all you need is a decent GPU (or the ability to cheaply rent one).

>>1986 >>1982 Please try using negative prompts. No reason to have mutilated watermarks in your generated images. Would be nice to have a few others actually contributing art in this thread. It takes very little actual technical skill, all you need is a decent GPU (or the ability to cheaply rent one).

>>1990 I've legitimately tried to see if I could produce a single decent output but even with an AI doing the bulk of the work my artistic ability is nothing short of comical. Even with your guys' direct help and taking a few prompts from the file names for guidance anything I produce can't hold a candle to your outputs.

>>1988 As the guy who requested the poodles in the first place, I can definitely get behind this, they can have big bellies and tight shiny swimsuits on the beach, here's hoping the A.I gods can make it a reality.

>>1989 I'm so used to signatures on art that the AI's smudge doesn't even register to me. Thanks for the reminder.

>>1991 Can you share some of your bad results? It's possible you need to use a higher step count, or are using the wrong model. There's a couple very simple things that make a big difference and are easy to fix. I assure you, it's completely possible.

>>1999 I've been in this thread repeatedly showing my bad outputs and procedures >>1856 >>1829 >>1497 step count is anywhere between 50-70 like >>1485 suggested, running out of ways to avoid the inevitable here.

>>2000 I know you can do this, and I know you've been trying. That was the whole point of this thread: to get other people using the software. Don't give up. You say you've been using two models at the same time. Have you tried just one? Have you played with the sampling or scale settings? Scale can make some difference, I haven't figured it out yet, but higher values seem to produce better outputs (up to a point, then they get worse). Are you doing img2img or txt2img? Negative prompts? Those don't show up in filenames, but can make a big difference. Stuff like "bad anatomy, bad face, ugly, fat, lumps, cartoon, lineart" etc etc. Anything you don't want showing up. What's your workflow? Do you generate images at a resolution and leave it? Do you run the outputs through the built in img2img or upscaler features? The unfortunate part about AI is that you will go through quite a few results before seeing anything good. It probably takes me 20 or so generations to get 5 reasonable pictures, and that's generating 8 images at a time. I've probably only gotten a total of maybe 20 really good, flawless images through the whole time I've been using the software, and I'm the one that started this thread. NovelAI is better but it also isn't free unfortunately (unless you want to track down the NAI furry model, if it's been leaked).

>>2002 > You say you've been using two models at the same time. Have you tried just one? Yes. No discernible difference. > Have you played with the sampling or scale settings? Tinkered with just about every setting I can tinker, that includes scale. > Are you doing img2img or txt2img? Aside from the inpainting attempt all txt2img, yet to produce anything that could be salvaged with img2img. > Negative prompts? Obvious ones like vore, obese, gore, may add more. > What's your workflow? Do you generate images at a resolution and leave it? Do you run the outputs through the built in img2img or upscaler features? As said before haven't produced an output that can be salvaged with img2img. Make 2-3 batches of 8, move some settings around, repeat.

>>2003 Alrighty. I have one more idea. Give me a scenario/character you'd like to see rendered in AI. I'll spend as much time as necessary building a prompt that gives a decent result. I will post that prompt, as well as every other setting I used, with the idea that you should also be able to get a good result on your own software. I'm actually interested to see how replicable AI results are. I'd also say there's a balance to be struck with the negative prompt. "Vore," for example, may be confusing it, because it's idea of a vore belly is very close to it's idea of pregnant. I've also never had to add "gore" to the negative prompts, just because it doesn't really recognize it. Try removing a few of those.

>>2003 I wouldn't use img2img to "salvage" results. I don't really "salvage" results, just move on if it isn't worth it. I use img2img to generate a few slight variations in a batch (noise scale changes how much img2img changes the original) on an image I already think is good, or can be good with a slight change. It can often reposition a belly button, improve a face, fix a hand, etc. Like I said, if you can give me a scenario you'd like to see, I'll give it as much time as needed to figure this out. This is an interesting problem and replicability is an interesting thing to explore.

>>2005 See if you can do the swimsuit poodles

found an AI artist on IB seems to focus on preg furries. mostly foxes

fox big

Is this thread dead? If not then i would like to request some otter girls

Thought you all might want to know about the discord community working actively on furry/scalie image generation. There's loads of custom models, channels for pregnant furries, etc. Some actually incredible stuff has been posted, and there's lots of discussion of an guides on promptcrafting, settings, etc. Invite code is hJVWeEkPPH

some things made with NovelAI

>>2170 >some things made with NovelAI can you explain the process for using novelAI for this sort of thing? I thought most of these AI tools were censored by their creators to not allow prompts that would produce lewds.

>>2173 >>2170 Seconded, please give us some prompts

Used Novel AI to get this. Prompt: wolf, arctic wolf, white fur, striped fur, striped markings, black stripes, tiger stripes, fat, chubby, big breasts, big butt, big thighs, big belly, nude, lactating, milk, black hair, blue eyes, multi breast, huge tail, fluffy tail, kitchen, kitchen appliance, kitchen utensils, Yes I used "fat" instead of "pregnant", but the result was better this time.

the prompts I used for mine on NovelAI are pregnant, girl, suggestive fluid, delivery, under night in-birth, breastfeeding, gaping pussy, showing nipples, completely nude, showing vagina, egg laying, crying, labor pain, fetal movement, pain,

>>2406 negative prompts?

Here's what I was able to do in inpainting on stable diffusion. I just filled in the belly and this was the first result. Although I can only make images in 512x512 I just took it to an image upscaler site to fix the poor quality.

>>2608 We need a lot more preg sted stuff.

>>2642 You got more of that?

These are the only other Amy ones ive got so far.

Those 3 amy pics are novel ai which is stable diffusion on steroids, like as if it had every character and every model all at once since its all connected no training required the training just comes from users using it, I would use it if it wasn't $20 a month

>>2608 Can you explain how you did this to me? Like, where the mask was drawn and what prompts you used, if any? I haven't managed to get anything but trash from inpainting.

>>2875 Okay, so, I've been going at this for like two hours now, and no matter what prompts I give and what settings I use, it always either generates random lines or thinks it's a fucking fishbowl or something. Man, I give up.

Maybe its the model your inpainting with or you have a setting too high

pregnancy lines are hard to inpaint tho, cause there aren't alot of models specifically trained for that or even know what it is (I guess)

>>1750 Could some one make that crappy belly i drew on there ages ago look better with the ai?

crappy inpaint

>>2902 Wow thats better then me! you think you could put at outie belly button on it too?

>>2903 that really sticks out? i forgot to add

>>2876 Did you make sure to set the inpainting mask to use "original"? The noise, blank, and fill options will obliterate the edit you've masked and make it unpredicatable

>>2912 Not sure if that's what I was using in that result specifically, but I definitely tried. Hell, I tried all the settings.

Good Lora's used with a good model and vae can do this.

>>3276 "Lora's" Let me guess, autocorrect? Or more like, auto cat rectal?

A Lora is like a hyper network or embedding I think but it uses the model good enough to where its as If it was trained in it idk

>>3292 Huh...well, today I learned "Lora" is not autocorrect for "Loona"...

I took someones preg AI and made them biger than everything. Have fun. https://beta.character.ai/post?post=1IAoYkaaYVRTnLtk05faxjW4zX86QJmFT_5BYtAIDX0&share=true

>>3422 Do you have the link for this chat bot?

Some of the better images I've gotten recently

MLP

>>3422 Link pls?

I dunno why FemShadow is so damn good for me, but Sonic Diffusion can sure do good sometimes.

>>3422 >>3425 Protip, it says who made the bot in the first message, so just look them up in the future. In this case, it's a fairly well-known artist/colorist who posted it publicly on Twitter. https://twitter.com/PacificSide18/status/1634663459320365056

>>3656 Thank you anon! I swear, the private bots must be at least ten times the number of the public ones, i really should start saving all the links i find in a .txt file

>>3657 I use a start.me page. You can check out mine if you want https://start.me/p/jjr9KQ/ai

>>3445 >sweaty u alright son

>>3658 Nice collection! Very much appreciated

Y this shit got me down so bad?

Ye

>>3763 Post the wrong file, did ya?

"Artist" is Infernal2 on FA but the image was LORA'd or something with BahnBahn's art as a training reference. Honestly pretty impressive apart from the tail.

Show me a knotted dick

>>3904 It was banned on FA eventually, but it moved to IB. https://inkbunny.net/ProjectSoulstealer

Anyone got any good results out of inpainting to make edits?

Went a little wild, had some fun with the curiosity. Learned a few things, and yeah. Might do more later if I can think of anything unique enough.

>>4755 Not too bad, I'd say.

Some selections from Inkbunny. Had to convert and compress to jpeg.

Pardon the realistic canine vagina...

>>4873 I see no reason to apologize, more of that please.

>>4874 you're welcome!

>>5027 Do some lizards or dolphins next.

>>5027 Man, wish I could get my stable diffusion to spit out results that detailed. What are the prompts you use?

>>5035 1 - Use model called YiffyMix 3.1 or 3.2 - they are the mix of Fluffyrock for most of the furry stuff + some of F222 for women bodies + other HD models -> therefore resulting in more anthro-ish figures leaning to realism. 2 - Also download embeddings named dfc, ubbp, updn, bwu, boring_e621_v4, and probably some others. 3 - use this prompt AFTER everything - (masterpiece, best quality, ultra realistic, 4k, 2k, (intricate:1.1), (high detail:1.3), film photography, soft focus, RAW photo, photorealistic, analog style, subsurface scattering, photorealism, absurd res:1.1) - you also may use keyword BREAK before all that, if you use Automatic1111 4 - the prompt itself is a free wording/tags, such as solo (anthro female SPECIES:1.2), lying, on back, pregnant, huge breasts, lactating, hands on belly + some keywords for background, such as detailed background, bedroom, bed, pillow, balloons, carpet 5 - use Hi-Res fix, especially good with 4x-UltraSharp upscaler beforehand. Ta-da! You are golden!

>>5034 Well of course. Here is some pre-releases that is yet to be loaded to Inkbunny :)

>>5044 Oooh these look amazing! are you open for suggestions? Because i would love to see some preggy Espeon art, not much of it to be honest haha

>>5047 Well, some things will come later (some models have troubles making good, realistic ferals). But better later than never, I guess. :)

>>5044 these are solid

>>5049 If you wish to see something more TRANSPARENT, you are welcome >:D

>>5048 Oh that's no problem! Even anthro is fine if it makes it easier for you

>>5050 I suppose it's worth asking in a general sense to those browsing this thread, but would it be too much to ask you (or me, since I'm dabbling in it on and off every now and then) to make suggestions for things? I am but a humble AI generator user who tends to feel very bored with his own ideas and just needs something to go off of every now and then, after all.

>>5053 Sure, How about a hugely preg semi-anthro otter?

>>5053 You can use some of mine suggestions. 1) Matatabi (naruto's two-tailed beast) 2) Meicoomon 3) Legiana (Monster Hunter) 4) Light Fury/Charizard Hybrid 5) Zoroark/Goodra Hybrid 6) For a harder challenge - any serpentine creature (Dragonair/Python/Naga/Master Viper/Eastern Dragon)

>>5050 the future is fuckin wild

>>5056 Probably not what you're looking for as far as the semi-anthro part is concerned, but got this out of it at any rate. >>5060 Pretty sure my checkpoints or loras don't know about 1-3, and any hybrids come out as unappealing abominations or something else. Any tips to reduce this? As for serpents, it's proven difficult to get an appealing pregnant serpentine that isn't just "snake but with an anthro pregnant belly." The AI can't seem to make up its mind on how serpents should look in that regard.

>>5071 Yep, I have a couple of tips for ya. 1) Related to picture: Eyes and some details are horrifying. To fix them, you need a process called Hi-Res Fixing. In short - you generate a picture, then you upscale it up to 2 times, and then you throw it into IMG2IMG with same settings you used to generate it AND denoising level ~0.4 to ~0.6 depending on how many details you wish to make or how many you wish to save. But for that process you may need ~5-7GB of VRAM. Yet, it will, for sure, give you much better results than usual. 2) To explore suggestions 1 to 3, you may need to learn how to make LoHA/LoKR/LoCON using Google Colab + Kohya Trainer. It's not exactly a nuclear rocket science, just google it and you will find exact methods :) 3) I create pleasantly looking hybrids by using Prompt Manipulation included in A1111 tool (and there is a node for ComfyUI as well). In short, if you use this syntax - [charizard|light fury], then at step 1, AI will attempt to draw charizard. At step 2, AI will attempt to draw lightfury on top what he is already drawn. At step 3, it will again draw charizard. At 4th it will change back to fury - and so on. You may hybridize more than 2 concepts, just write them as [A|B|C], or [A|B|C|D], heck, even [A|B|B|A] works too and gives awesome results as well. 4) Well, thats why I called it 'harder challenge'. I stuck at that part too :)

>>5056 >>5071 Here is my take on that otter. If you wish to know how I've done that, look at my workflows (I use ComfyUI tool with some custom nodes, every bit of information included with the workflows) : https://pixeldrain.com/u/9ZdNK1Dw

>>5086 Password on archive is : 321 (sorry, forgot to mention)

>>5036 Trying to follow your instructions and in typical fashion I immediately hit a brick wall with step 1. Downloading the model from Civit and it does not show in my SD UI checkpoints

>>5088 1. Make sure you placed model in correct folder 2. Make sure it is a safetensors model, and has at least 2GB in size 3. Try to ensure that model name have no spaces or special symbols in it (for example, let it have name yiffymix_32 instead of "Yiffy %Mix #3.2") 3. Restart your preferred software to ensure that it loads with that model in folder.

>>5089 1. It's already in stable-diffusion-webui/models/Stable-diffusion, Civit says put it in a VAE folder but no such folder exists. 2/3. Downloaded as yiffymix_31.safetensors, 2.082 GB 4. I'm not *that* inept to have not already tried restarting after downloading it

>>5090 Hm... it does sound correctly, however, then, there is more questions. For example, question - only NEW model doesn't show up, or ANY model? In case of the latter, perhaps some settings got jumbled up (like home path, or path to models folder). Also, there is models/VAE folder, but I don't think it should be relevant at all, since it is where you store VAE models (300 to 800 megabytes in size), not drawing models.

>>5091 > only NEW model doesn't show up, or ANY model? Only the new model, other models I've used during my (pitiful) attempts in the past still show up. I found a download for the A1111 you use which may help me through at least this wall.

>>5092 Interesting... this MAY be caused then by literally ANCIENT A1111 version. When you updated it last time? Because it is started a long time ago as being only .CKPT program then. If not, then if you are at linux version, you might have wrong permissions set to a model, so it became unreadable by the user who ran A1111. It is really funny and interesting problem, never encountered that one.

>>5093 I thought I had updated it before testing but I was dead wrong, now I've got it at least acknowledging the model. Now to maybe produce something not embarrassing.

>>5086 I’m curious as to whether or not you could make her even bigger?

>>5096 Of course is possible. But beyond certain threshold it starts to lose to more balanced versions.

>>5097 I gotta give you credit, your stuff (while not perfect, such as with the two navels here) by and large looks pretty good.

>>5098 Happens, when you disbalance picture in favor of a specific tag. Without cleanup it was significantly worse, with broken paw, missing tail and 4 navels. I was just pretty impatient when making this one, as I could have waited till more variants were generated to be composed together.

>>5085 Definitely some curious things to consider going forward. The eyes before up-scaling looked fine originally, but I suppose it's worth looking into in my downtime for fine-tuning. Also I suppose it's worth asking, but are there any other syntax modifiers that can be used? On another note, I've been trying to force certain camera views or body rotations, but I only ever getting the one frontal/partial frontal and it's getting rather old to me with the little variance I'm getting with the prompts I've been using.

>>5099 I realize, rereading my comment, that that kind of reads like a backhanded compliment. Sorry, not how I meant it, I get it happens with AI stuff. But really, your stuff does stand out from usual AI in a good way.

>>5100 Yes, there is a couple of another syntaxes, like [A:B:0.7] winch will draw A for the first 70% of image and B for the rest 30%, or [A::0.4], winch draw A for 40% and 'nothing' for other 60%, or [:B:12] that will draw 'nothing' until 12th step, and then start drawing B to the rest of the picture. To modify camera views, you most certainly want to slap down some view tags - such as - 'rear view', 'front view', 'first person view', 'side view', 'low-angle view', 'three-quarter view', 'high-angle view', 'worm's-eye view', 'bird's-eye view', 'waterline view'. If it is not what you desire, simply increase emphasis on the view by 10-20%. (side view:1.2)

I am usually much more active @ Inkbunny ( https://inkbunny.net/ProjectSoulstealer ), but I think of dropping my domain somewhere here with time too. With less of a 'project' and more of a 'soulstealer' aspects, so to say. This is going to be great! :)

>>5113 Interesting. I've been getting some decent results because of it, though there is a new issue that I don't think can be solved with syntax. It's also come to my attention while looking through most of the pictures here and I've noticed a glaring discrepancy, but it seems to me anyway that there isn't a way to get the AI models to register fetal movement and/or bumps. I'd been playing with the prospect for a few days now but with no luck on the matter. It's likely a matter of getting the right prompts with the right checkpoints and so forth.

>>5146 Not really. You can get a lot of bumps. There is a couple of techniques to do exactly that. 1 - before you do high-res fix, open your image in any photomanipulation tool. Create new layer and set its opacity to 10%. Get either a black or a darker-fur-color brush, and set your brush to have 100% hardness. press a round brush somewhere. Take eraser and erase smaller circle inside, so result would look like crecent. Repeat it 5-10 times. Throw it into IMG2IMG, but now instead of copying your prompt from last success, instead, append 'baby bumps' or any other similar tag to it. In 3 to 10 variants you will get what you seek. 2 - train LORA on images with a lot of belly bumps. 3 - in worst case scenario, use tags such as 'vore'. You can't really tell if picture is about eating someone alive or carrying one, that's only determined by tags usually.

How is this