>>8922 Images that have been processed with hi-res fix or detailing, while typically imperfect, are preferred.

>>8922 For those of you that are asking "How can I do this myself?" There are many answers! My preferred method is a custom ComfyUI instance running on lykos.ai (the local app). NovelAI provides a decent service, but will cost you a bit. If you don't have a good computer or money, google colab offers free usage of relatively high-power GPUs. I have not tried this, and cannot provide advice. The last thread had a lot more advice and recommendations, unfortunately it was lost.

And in the spirit of things, I'll share a few of my retouched (hand-edited) pics from the last thread (and a few that never got posted!)

Anyone still have the loona xeno preg ai art saved?

>>8933 great find! I also remembered there was a tigress version too

Hopefully the quality of these two are good enough.

>>8937 Not bad, but I might tweak the eyes/mouth on Krystal a bit.

>>8934 >>8935 Here's a continuation to the original Tigress pieces. I enjoyed the style/premise the first time, so I figured I'd stick with it for the follow-up. Seems as though it's been a couple of days and she's really starting to feel the effects of it... Looks like she could use some lotion with all that rapid growth, too.

>>8939 And an internal version as well. If there's any input to be had on the transparency/overlay, I'd appreciate it.

A recently finished piece that had been sitting in my "unfinished" folder for quite some time.

>>8938 It's always the damn mouth and the eyes that keep sneaking under my nose.

>>8939 One thing is for sure. It's not going to come out of her vagina. But maybe though her butt?

>>8942 ComfyUI offers the ability to do small, regional tune-ups. There are some face recognition models geared for furries too. You can set it up to automatically refine the face/hands - not always better, but usually

>>8940 >>8939 kino

>>8948 Pregnant Cynder looks great and awesome, but can you please generate pregnant anthro female dragon Spyro?

>>8951 Why not? Here's three different styles and poses. I didn't feel like touching any of them up, so they're not perfect, but not bad either.

>>8951 Aaand the last one.

>>8951 cause that stuff goes in the r63/tg thread.

>>8961 Oh yes! Thanks!

>>8961 Hey mister , i love your work , you can make a good tutorial to use comfyUI?

>>8964 I'm working on a written tutorial. No assurances it'll be useful, but theres a lot of online resources available for this sort of thing. I'm trying to narrow in on how I use ComfyUI specifically.

>>8967 Not a fan of MPreg unfortunately. You'll want to try ComfyUI yourself, or head over to the MPreg AI thread and see if one of them will help.

>>8964 Thanks! Unfortunately, I'm not a videographer or writer by trade, so I can't assure you that this tutorial will be "good." It is, however, how I achieve my results. If you're going to use my methods, you'll need a computer with decent-ish specs. Mine has an RTX 3060, but I don't think you need something high-end if you're fine with waiting a bit longer. You will no doubt run into problems trying to get this working. It's not as easy as something like NovelAI, but it is free, and it is powerful. I ran into many hurdles where things just didn't work for reasons I didn't fully understand. Keep trying! I promise, it's possible. A lot of other people have good tutorials on how to fix some issues. Others, you'll have to figure out yourself. You'll want to use lykos.ai's "stability matrix" application. It's a handy wrapper for stable diffusion and ComfyUI. This part is pretty basic, so you should be able to find a youtube tutorial for running ComfyUI via stability matrix (that's what I did, after all). Once you've got the application set up and running, you'll want to install the ComfyUI manager. This will help a lot in later steps. This is another place where you'd be better off looking up "how to install ComfyUI manager" than listening to me. Now you can launch ComfyUI! It'll spray out a bunch of console stuff (i'm not a computer scientist either), and eventually "To see the GUI go to: [local server link]." Click the link, and this will open up the ComfyUI instance in your browser. Don't worry, it's a purely local instance - no-one else can connect (although if you wanted to, you could set that up). ComfyUI is a sort of backend tool. It's got a plug-and-play system where YOU need to place the nodes to load the model, encode the prompt, and sample the image. This has upsides and downsides - a major upside, for example, is the fact that you can install custom nodes (which is why we installed ComfyUI manager earlier) to do things like refine faces or hands automatically. I prefer the EasyFluff V11.2 model. You can find this on Huggingface, posted by Zatochu. Find the "EasyFluffV11.2.yaml" (config) and "EasyFluffV11.2.safetensors" (model) files in the file tab, and download them. You'll want to put these both in your "stablediffusion" folder, which you can find by going to the "Checkpoints" tab in StabilityMatrix and clicking "models folder" in the top right. Pay attention to the model card - you'll need to install the "sampler_rescaleCFG.py" node to use EasyFluff properly. Here is where things get a little difficult. You'll want to install "ComfyUI Impact Pack", "ComfyUI Inspire Pack", and "ComfyUI Prompt Control" via the ComfyUI manager. If you can't figure out how to do this, this is another step you can probably find on youtube. Once you've got EasyFluff and the custom nodes installed, you'll want to pick a blank spot on the ComfyUI canvas (or make one), right click, and click "add node > advanced > Load Checkpoint with Config (DEPRECATED)" This is necessary to run EasyFluff properly. Select EasyFluffV11.2 for both the config_name and ckpt_name fields. Okay, so now if we click "queue prompt" we'll successfully load the model - just not use it yet. You'll notice that the checkpoint loader node comes with three outputs - "MODEL", "CLIP", and "VAE". To use the model, you'll want to click and drag these outputs to the appropriate inputs on other nodes. Create a new "RescaleCFG" node through "add node>advanced>model>RescaleCFG". Drag the "model" output from the loader to the "model" input on the "RescaleCFG" node. Set the multiplier to something between 0.70 and 0.80. Next, you'll want to create the prompting nodes. "add node>conditioning>CLIP Text Encode". Make two. One will serve as the positive prompt, and one as the negative. These nodes take "CLIP" input from the loader, and output "Conditioning". One last step before we can sample (generate the image!!!). "add node>latent>empty latent image". This is where you can set the size of the image to be generated, as well as set up multiple batches to be run if you so choose. "add node>sampling>KSampler". You'll notice "model" "positive" "negative" and "latent_image" inputs. Link these up with the appropriate outputs from your other nodes. The sampler will show you a preview of your image, but this isn't final. I find that 25 steps with 7.5 cfg on euler_ancestral sampler with a normal scheduler and a denoising of 1.00 works well for EasyFluff. Now, when you click "queue prompt", it will take the input from your model and text encoding and generate an image with the sampler. You still need to decode the latent image though. "add node>latent>VAE decode". This will take the latent output from your sampler, and the VAE output from your model loader, and output an image!. I like to upscale the image by 2x before displaying it with "add node>image>preview image". At this stage, you should have a functional image generator where you can input a prompt and get out an image. I prefer EasyFluff specifically for its ability to adhere to artist styles, avoiding the usual "hi-res fixed AI image" look that a lot of AI products have. This gives you a lot more control over how you want the final product to look. Like a painting? OK! Like your favorite artist on E621? Sure! (works best if the artist has 100+ tags). A wholly unique style formed from a composite of one or more wildly different artists? Absolutely! I can't really explain well how to piece together the automatic detailing functions. It took me a lot of time, and at this point I'm just using my past work - not really "understanding" it. Luckily for you, you can use this past work too! If you've figured out ComfyUI well enough, and have installed all the custom nodes (including bbox FacesV1 (the furry face identification model) and the other default bbox models) then you can simply download my ComfyUI workspace and use it! It's messy, complex, and hard to figure out, but it is functional, and it includes a lot of separated/grouped nodes with brief annotations to (try) to explain things. It also includes a lot of artists that I've found to work well in prompts, as well as "default" negative and positive prompts to base your work on. If you make it this far and would like to try using my workspace, let me know. Otherwise, if you have any questions or feel as though you need assistance, I may (or may not) be able to help. This took a lot of trial and error on my end, and got frustrating on many separate occasions. Keep at it!

>>8970 I can also offer further advice on prompting, image editing, and settings.

WHAT THE FUCK DO YOU MEAN THIS SHIT IS AI GENERATED THIS IS SO GOOD

>>8973 EasyFluff is in my opinion an abnormally good model, but it also takes some knowledge on how to use it. The vast, vast majority of AI "content" (eugh, hate that word) on the internet is made by people with no understanding of the underlying process and no real eye for art - hence why you tend to see people posting massive batches of mostly slop. You have to understand that for each of these "good" images, there's a solid 50-100 that are discarded for some weirdness or abnormality. I spend a lot of time examining the image as it's being sampled, and cancelling it/moving on to the next seed if there's something obviously wrong. Sometimes I make it past the first step, and the second, and correct the face, and hires fix it, (a solid 30 minutes to an hour of time) only to discard it when the final product just wont turn out decently. Even then, these aren't perfect. (but then again, neither are most artists). I also avoid certain subjects and poses that just don't work well. The datasets that AI is trained on don't include hundreds of literally every type of subject, pose, and style - just most of them. It's an active process trying to stay inside the boundaries of what AI can do "well."

>>8973 You can also, of course, thank the artists who have gone through the years of training necessary to produce good work, posting hundreds of images to E621 and other art accumulation sites. These people are the core of my work, and I wouldn't have been able to do what I do without their effort. Unfortunately, in a lot of cases, these artists have their work scraped without permission. I like to think this isn't replacing anyone, merely opening up the playing field. Many of my inspirations just don't draw pregnant dragons - but models assembled on their work and the work of others can.

so... can anyone generate Renamon, Gardervoir and Krystal Fox?

>>8985 I think that can be arranged. Any ideas for context/fantasy? I'm not terribly familiar with any of the source materials, so it would help to have a bit of context for the characters. Plus, it'd be fun!

>>8986 Well, i too don't know much about the characters, all i know is Renamon is from Digimon and Krystal is from Star Fox (cmiiw). as for Gardevoir she come from Pokemon, shes a psychic type pokemon. that all i know, perhaps someone could provide more.

>>8986 Could i also request Anthro Rainbow Dash, Twilight Sparkle and Apple Jack too?

>>8988 I'll get around to it, but in the meantime, have these. Generating ponies is interesting because there's about 10/20x more pony content in a given training set than there is for any other specific character. Look at the number of tags for rainbow dash on e621, then look at renamon... I usually use EasyFluff, but there's a specialized model trained on Derpibooru material called Pony Diffusion. Its easier to run - if you're interested. It still struggles with the cutie marks though. Another bit of trivia, but the (now lost) original AI thread actually began with Rainbow Dash!

I'll repost some of AdobeWTF's best Luna generations. I'll begin with the muscular Lunas.

>>8990 Excellent! If I remember correctly, these were made with Pony Diffusion. Luna's hair is really pretty - and that particular model captures it well. My attempts with EasyFluff were a bit... flat.

>>8990 These are a little weird, but I still like them. I downloaded all of his generations from the last thread, and can upload them somewhere in a zip if anyone wants.

>>8990 >>8992 Here's my take on Luna (via EasyFluff). You can see it doesn't get the hair or cutie mark quite right. I had intended to edit out the second horn, but I never got around to it.

>>8985 Here's a few of my unrefined generations from way back when. I'm also a renamon fan, so I'm putting a little more effort into a special image as well, but why not post these too?

>>8993 Yeah, I saved yours too.

>>8995 Damn, I didn't even have that one saved. We'll never know why either, cause it's not half bad. Thanks!

>>8985 Here's Renamon. She's digital, so I figured I might try that new "digimon controller" app to play with her a bit. She thinks it's for fighting... Let me know what you guys think of the "annotations." I put a decent amount of effort into the composite pieces, but some are still better than others.

>>8985 Turns out Krystal's a telepath. She's spent a lot of time out and about with Fox over the years - and a lot of time listening to his inner thoughts. When Fox told her he "liked her like this" ... Well, let's just say it wasn't a surprise.

>>8994 >>8997 >>8999 ooo these are nice, Thanks! i'll be waiting for gardevoir next

>>8970 Can you show screenshots of where to put EasyFluffV11.2's and the CFG Rescale where in the StabilityMatrix folder? My brain is about to explode because I can't figure it out on my own.

>>9008 Sure! You just need to find the "models" folder. You can get to this through the path shown in "guide1" or by navigating to the "checkpoints" tab of stabilitymatrix and clicking on "models folder" in the top right.

>>9008 This is the path to the custom nodes folder for ComfyUI. I'm a little confused myself, because my folder doesn't include the rescaleCFG node, but all my other customs are there so I'm *assuming* it's the right place. Like I said, this took a lot of trial and error for me. Good luck!

>>9019 >>9022 Thank you~

I got an error for generating AI art(s). ~ Error occurred when executing KSampler: CUDA error: the launch timed out and was terminated CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

>>9024 I googled the question and found this potential solution. hxxps://www.reddit.com/r/comfyui/comments/17pxrau/please_help_vaedecode_cuda_error/?rdt=41873 (copy and paste the link, and replace "hxxps" at the start of the link with "https". This avoids the website getting analytics saying it was visited from Pregchan.) I ran into a lot of similar errors, and it's likely you'll run into more. Keep at it!

For funsies, a Kobold and her Master (and their child).

>>8988 Booted up PonyDiffusion for this one. Great model, does well with the style Loras and some pretty complex prompts.

>>9034 I might touch these up at a later date. There's a few small errors that could be cleared up pretty quickly by hand (face details, clothing, and the hair by Twilight's feet).

I have a hard time to fix this issue. I have to add a spoiler because my positive prompt is mpreg based

>>9046 You'll want to check the stability matrix console. It's impossible to tell what's gone wrong here with no debug info. You'll also probably want to use a "preview image" node and not a "save image" one. That lets you inspect the result *before* saving it (by opening the image and saving it manually).

>>9046 Took another look. You're generating your latent image at a CRAZY resolution. Not possible to generate an image at that scale - it's probably this causing errors. I stick with 768x1128 for my max usually. If you want a higher resolution image, you'll want to scale up the output. You can either use a plain upscale, or upscale it and run it through a sampler at a low denoising value (0.2-0.5) to perform a hires fix. Image generation models are typically trained on image sets with a specific resolution. EasyFluff is trained on 1024*1024 if I'm remembering correctly. Generating large images just won't work.

>>9049 >>9051 I got a VAE Decode error here.

Do you guys have a list of negative prompts?

>>9054 Can't help without a little more info about the specific issue. I'd probably just Google the error code though. Check the console in the "launch" tab of StabilityMatrix. I'd suggest trying an image gen at 768*768 and moving up the scale from there maybe? You'll also want to include a couple artists at the end of your prompt when you get the error figured out.

>>9056 CUDA error: the launch timed out and was terminated CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

>>9056 >You'll also want to include a couple artists at the end of your prompt when you get the error figured out. Any artists like bahnbahn (artist:bahnbahn), just for curiousity

>>9058 Basically any artist you can find on E621 with more than about 100 posts. Format for artist tagging for EasyFluff is "by bahnbahn" not "artist:bahnbahn". Generally, using more than one artist works better, and the order of the names will influence the output significantly.

I have to always restart my comfyui every time I get errors.

>>9060 Have you changed the resolution of your latent image? It should be much lower. My instincts are telling me this is a VRam related issue. If you lowered the resolution and are still running into problems, you may want to enable "--lowvram" in the launch options (settings gear by the launch button in StabilityMatrix app).

>>9055 My default negative prompt for Easyfluff is: lowres, bad anatomy, bad hands, text, error, cropped, This changes depending on the specific needs of the image. Generating teeth where I don't want them? add "teeth". Hips a little unnatural? add "hourglass figure". That dragon looks a little fuzzy? Add "fur" or "hair". Et cetera.

>>9034 Say, want to give Rarity a try, or attempt to generate some fat-assed Celestia?

I may as well post some more Lunas, these jogging, and this Chrysalis I found.

>>9066 Sure! Any preference on style?

>>9068 These were pretty good: >>9034 The Rainbow Dash in particular is damn near perfect.

Hey did anyone save any of the images of the pregnant cats from the last thread before it got deleted in that purge?

>>9066 Here's one of the better generations of Rarity. I've got one that I'm gonna retouch, but this one didn't turn out bad - just kinda weird feet.

>>9072 Thanks. This is perhaps the best one yet. How long and how much did you work on it?

>>9074 Maybe 30-45 minutes? PonyDiffusion works pretty well, but I took some time to narrow down the prompt.

>>9081 If the hair comes across properly, Celestia's gonna be real good.

We've come a long way from the monstrosities of Stable Diffusion 1.0

>>9066 And a retouched version

>>9081 Hey, thanks again. These are arguably even better. >>9083 >>9084 >>9085 Yeah, these are surprisingly good. I've been critical of image generation models, but I can't argue with results. I may end up using these things to make cover art for some of my writing in the future.

>>9086 That'd be interesting to see. As always, I'm willing to share any advice/resources I have. I wrote the tutorial for EasyFluff, but everything I've posted since >>9034 were all PonyDiffusion. Just slightly different.

I have had a groundbreaking idea that may result in some very interesting results. My setup allows for the hi-res fix to use a *different model* than the original image generation, as well as an entirely separate prompt. I have not yet made use of this, but... I tend to like the body shapes that come out of EasyFluff more - they're a little less conventional, and more within my control. I also like the way some artists draw certain bodies. I also, however, really like the style and vibrancy offered by PonyDiffusion. Generate an image via EasyFluff, then hires fix with Pony Diffusion?

>>9056 >I'd suggest trying an image gen at 768*768 and moving up the scale from there maybe? I set to 768x768, but I still get errors on the way out. I don't know by "moving up the scale" you are saying.

ComfyUI has updated recently. So, I updated it today.

>>9093 I thought that would fix it and you could increase the image size from there (hence, moving up the scale). I'm afraid there's not much more I can do to help, you'll have to troubleshoot it yourself. What is the VRam of your GPU? Did you try it with "--lowvram" ?

>>9097 1. I have 1995 MB of VRAM 2.Yes, I set to --lowvram, but still.

I'm getting annoyed by the "RuntimeError: CUDA error: the launch timed out and was terminated" It drives me nuts. I can't troubleshoot it yourself because I'm very good to fix things. I have "--lowvram" selected for ComfyUI via the Stability Matrix software, but I'm still getting the same error over and over.

>>9099 >I'm very good to fix things I meant to say that I'm NOT very good at fix things.

>>9087 >That'd be interesting to see. As always, I'm willing to share any advice/resources I have. Oh, I don't have any of the hardware for this, so it would be years from now, if at all. For now, I don't want to be greedy and ask you for anything else, unless you'd be intrigued. I've got an anthro hyper pregnancy story that seems like a decent candidate for this kind of cover art. On that note, have you ever generated hyper pregnancy with these models? How do they handle it, if so? >>9088 >My setup allows for the hi-res fix to use a *different model* than the original image generation, as well as an entirely separate prompt. Go for it; I've seen all sorts of cool shit enabled by using different models for different stages.

She is my "oc" Shirley a anthro lioness teacher

Heya, just a little reminder for using ComfyUI: You are allowed to generate AI arts from 512*512 and nothing else. If you go higher with 768*768 or 1024*1024, you will getting errors.

>>9115 This is a consequence of low VRam. I have 8gb of VRam. I can generate images at 768*1128, 1024*1024, etc. When sampling an image, your computer has to be able to "look" at the whole thing at once to adjust it, so it stores it in VRam. If you have less VRam available, your computer can "look" at less of the image at once, which makes the parts it's not "looking" at come out strange. If you try to "look" at too big an image all at once, you crash. I'm glad you got it figured out though. I'm sure the MPred thread will appreciate it. A couple other of my favorite artists to use in prompts in no particular order, Oouna, Photonoko, Taranima, Vale-City, Panken.

I only have 2048 MB of dedicated video memory, or 1995 MB of VRAM. I don't think I should make bigger AI arts, sorry.

>>9122 Not bad! >>9123 Don't worry, there's still things you can do to get better looking pieces! You can upscale your images with an "add node>images>upscaling>upscale image by" node. Easy, and computationally cheap. I like to upscale my images by x2. If you're interested, you can also refine the faces automatically. This takes a much smaller part of the original image and inpaints it at a higher resolution (if you do upscale it). Smaller than trying to generate one massive image, so it might work on your system. If you've got the ComfyUI Impact Pack installed, you can right click, "add node>impact pack>simple>FaceDetailer". Simply pipe the correct nodes in from the Impact Pack, selecting appropriate models. I've attached a screenshot of my face detailer for convenience, all the appropriate nodes can be found in ">impactpack". The default detector models are pretty shitty at picking up anthro faces, but you can find "furry face finder" models on line. Should show up as a .pt file. The path to place one in when you find it is "StabilityMatrix-win-x64\Data\Packages\ComfyUI\models\ultralytics\bbox"

You think the programs you use would be able to make preggo mamagens?

Wanted a more detailed background from EasyFluff, so I added inpainting/img2img to my setup and went for it. This opens up opportunities for more detailed settings and contexts. First I took an image of a train station and img2img'd it with the style I wanted. Then I made a dragon (just to test, of course), and placed them in the image. Then I took the combined image and processed it again, then went back in and added face/hand details manually. I suspect this could be put to tremendous use by taking screenshots from games to use as a background. You'd get to choose your own camera position, angle, height, setting, etc - then transform it.

>>9128 You mean protogen species? Yeah, I don't think that'd be much trouble.

>>9129 Here's the dragon before she got smushed down. I'll figure out the loss of detail eventually.

>>9128 Looks like these two friends aren't quite on the same page about what those pills were supposed to do...

>>9132 I had a lot of fun doing this one. Protogens are cool.

>>9101 >On that note, have you ever generated hyper pregnancy? Yeah, they can do it. It takes some extra effort, and its been a while since I've done it, but EasyFluff can handle that sort of thing. I doubt it can really get much larger than this though - at some point it'll stop recognizing the body as a body. I suspect it'd be a little easier now than when I originally tried it. Feel free to shoot me the idea. This site doesn't have DMs, but I'm sure we can figure something out. No discord or any other service that requires more than an email to log in.

my sona

Dug through /trash/'s Stable Diffusion threads a little. (Spoiled for possible cub.)

Had to convert most of these to jpg.

Spoilers for cub and possible unbirth?

>>8922 I appreciate the hell out of your initial description. Anyone who edits hands are god sultan hercules king of AI generation. The hands thing distracts me so much, I instantly skip AI posts that are otherwise outstanding.

Why do I suddenly want to see a anthro rottweiler pregnant with a xenomorph?

>>9132 mamagen (prompt)? o.o

>>9127 Do you have a workflow?

>>9148 >>9149 >>9150 >>9151 >>9152 >>9153 Okay. These are pretty damn good, I'm impressed. I've been trying to figure this out since the last thread, but if ANYONE has some insight on the artists/styles/prompt used to achieve that style on... >the hyena/possum from >>9151, >the hyenas from >>9152, and >the fox/leopard from >>9153, Well, I'd be happy to learn. And my posts would certainly benefit from a couple new styles...

>>9157 Unfortunately, I've discarded my original prompts from the protogens. I will say that I didn't get the image all in one go, and that I generated each protogen and the background separately. You will, however, find great success with variations on: anthro female protogen, pink fur, pink body, big breasts, pregnant, by bahnbahn, by panken, by oouna, And an example, because why not? I did remove an extra navel, and tweak the eye a bit, but aside from that this is all-AI.

>>9159 Here's my workflow. ComfyUI encodes the full workflow (wow!) in each image if you just click "save image". I've done that for a protogen generation and attached the file here, unedited. To run the workflow for this image on your own machine, simply download the image and drag it onto your own ComfyUI canvas. Wow! This *should* work, but if it doesn't, let me know and I'll find another way. It's a little confusing, but a damn sight better than it once was. Let me know if you manage to get it running and have questions.

>>9156 Hmmmm....

>>9165 Extra belly button as negative prompt.

>>9167 I find that it has a hard time recognizing what "extra" means, and it's a lot easier to just skip bad generations or edit it manually, which is what I did with the final version of that image. The unfinished one is just here to serve as a workflow download. This makes sense if you think about the training data - not a lot of images have been tagged "extra belly button" so the words mean effectively nothing. "navel" is the term used in e621 tagging anyways if I'm remembering correctly.

>>9160 >if ANYONE has some insight on the artists/styles/prompt used to achieve that style The regulars on /trash/'s /sdg/ share their prompts, models, and loras pretty often. You'll recognize the prompters' styles when they pop up, and then you can ask the poster to reveal their secrets.

>>9170 Wonderful! I'll give that a shot, thank you.

>>9134 Alright. The story's already published, so here it is: https://www.fimfiction.net/story/522544/preposterously-pregnant-pony-problems I don't commission anthro, but I was going to make an exception for Vale-City; unfortunately, I didn't snag a slot from him. These models probably don't cope with multiple characters well either, do they? There's a scene in particular I thought could make nice cover art, and it's easy to describe to a person, but I don't know if these things will do a decent job with it. >I'm sure we can figure something out If you'd want something in return, about the only thing I could offer is a very short story.

>>9170 My god. I've been there two minutes and already picked up so much. People are doing COMMISSIONS with this shit? A bit immoral if you ask me, but 40 dollars is 40 dollars.

>>9158 Cynder looks nice, but can you please make a pregnant fem Spyro too?

>>9183 perhaps in quantities of several pieces, if possible, maybe Spyro birthing

I generated two failed AI arts of Pom Pom from Super Mario 3D Land/World. The first one I failed is a red wolf with mushroon on her head, and the second one I tried is to stick closer to the prompt from e621, but instead, I ended up with a female Bowser, with a black wolf nose.

An anthro female Spyro the Dragon, pregnant, pole dancing while shoving her vagina on a dance pole. WTF

>>9190 Oh yes! That's wonderful! So exciting! Keep up the good work sir

>>9180 I have a lot of studying up to do, and a new VPred (same type as EasyFluff!) model to figure out. Very exciting. I may also try doing my HiRes fix at a higher resolution. The people at /trash/ /sdg/ are showing me up with their damn fancy machines with 32+gb ram. I'll post results when I have them, but I need a break for now.

>>9191 Dude, you don't understand: Spyro is shoving her pussy on a dance pole.

>>9192 >I may also try doing my HiRes fix at a higher resolution. How can you do that?

>>9195 I have a 3060 laptop GPU with (fairly sure) 8gb of VRam. This lets me do HiRes fix at twice the starting resolution of 768*1152. I might be able to pull of a slightly higher resolution. Maybe I just don't have the tech. If you're asking how to HiRes fix, check my workflow in >>9165. You take the output image, upscale it, use the "VAE encode" node to put it back into the latent space, and run it through another sampler.

>>9190 Some models haven't fully grasped poledancing, there's a second, disjointed pole behind Spyro's head. Pic related as an quick attempt with a model I use, the pole is slightly off.

>>9198 Derp, helps to actually attach the pic

>>9202 Heaven yeah! Fem preg Spyro are incredible nice and awesome, that's what I'm talkin' about, that's it! You make my dream come true, cured my longtime depression, you've done me proud, sir! Thank you so much, cheers!

I got my hands on a big massive list of 1900 artist tags and i've been sorting through them. Definitely a few standouts so far.

>>9212 Same image after HiRes fix.

>>9031 thank you so much!

>>9213 These images are insane, but I can't think of a furry character i could request. Do you also accept request from /d/ too?

>>9215 Depends. The models I use are tuned for furries, so the closer the style, the better. I'll go out on a limb and say that anime will flatly not work. It will handle a wiiiideee variety of anthro characters though.

>>9216 I see, let's try with french cartoon Wakfu then, in this case Miranda (she's an Ecaflip, one of furry class/race from the franchise and also canonically pregnant in the show and comic), since so far no one can generate Evangelyne close enough (but i guess you could try her too).

>>9212 Whoever suggested the protogens was batting on a different average that day. That bed is gonna be real messy by the time they finish.

>>9217 Here's my best shot at Miranda. It's not easy to get specific clothes. Good-looking clothes? fine. Specific ones? No.

Here's two AI arts of Lady Britven of the Bottle (anthro female blue sergal, pregnant) and one for Gleaming Shield, (anthro female shining armor, pregnant).

gentle reminder that the FaceDetailer module for ComfyUI exists and will turn a 3/10 image into a 7/10 one (at minimum).

>>9258 Hmmm... How does it works for Spyro, that's interesting

Can chilli heeler be recognized by the programs? If so can you make hyper preg art of her?

>>9269 Not bad!

Are there any models that would be capable of generating muscular pregnant females with muscle definition on the belly? Something like this pic

>>9275 Try indigomix. Prompting for it is a little finicky, but you might be able to find a lora that works for it?

>>9276 Thank you for your suggestion. This picture is actually really good. If there are any more results of this kind, I would appreciate greatly if anyone posts them. They are quite rare in traditional art, so at least perhaps AI can draw more of them. If possible, I also would like to know which model or lora might be used for other types of added features like stretchmarks, kicks or veins.

>>9298 Never thought I would enjoyed seeing these, quite cute.

>>9298 Yeah, not bad at all. Got a little soul to 'em.

>>9298 Huh, so ai knows who Sif is?

>>9034 Something I really like about this Rainbow Dash is the little red star right next to her belly. It's the only red star in the image, so it's kind of like her baby is cheering with her, just a little, just as much as a baby can.

This is my oc Bella