Hello everyone, I'm Bleahboy. Maybe you've seen my shit on Eka's or DA. It's not good don't go looking. If the mods will allow it, I would like to make this thread about how to make AI generated vore images. It may be an unusual style of thread but since there's really no other place for a post like this I figured here it would be ok. If the mods nuke it, oh well I guess. Before we begin I would like to explain a few things. AI can be used to make a LOT of different images and it is a powerful tool. With that power comes a price. To generate images on your own computer you will need to meet the minimum hardware requirements and have the minimum amount of autism necessary to learn a new shit skill like this. It will take time and practice and a lot of trial-and-error. You will see shit that makes no sense and be haunted by some of the things you will generate. Don't be discouraged, endless gooning awaits you. So here we go In the benninging, there were guides written by other people. https://aienthusiastic.com/automatic1111-guide/ Follow the above guide for how to install a program called Automatic1111. It is what you will use to generate images. A1111 uses your own computer to generate images and is completely free. Though it may be mining Etherium in the background idk. Some people love it, some people hate it. I use it because I have it. The things you should be focusing on in the guide is mostly the installation and the basics of generating. Don't worry about any of the Img2Img stuff, I literally never use it. If you want to use it, it will still be there after you have understood the basics. What we are focusing on is Text To Image (t2i), which is where you will be spending most of your effort. Simply input what you want in the prompt field and the AI will try to make something based on that. This is the most basic level of generation, we will go much deeper here. After you have been thoroughly dissatisfied with your results from the default systems in vanilla A1111, it's time to juice things up. A1111 is a powerful tool because it's open source, so a lot of other people have already done the hard work for you. To generate images that we actually will like we need to use "checkpoints" and "LORAs" that have been provided by other people on https://civitai.com . Checkpoints and LORAs are the brains and the instructions of A1111. The checkpoint will greatly influence the overall look and style of the image, while a LORA will affect what is happening in the image. A single checkpoint (usually) is used when generating an image, but, much to our delight, we can use many LORAs at once to craft bizarre and unique fetish porn for ourselves. We're going to download some LORAs and a Checkpoint so we can recreate an image and get started with generation proper. Download these things: https://civitai.com/models/257749/pony-diffusion-v6-xl Place in your A1111 folder in "models" > "stable-diffusion" https://civitai.com/models/589187/clementine-the-walking-dead-season-2-sdxl-pony https://civitai.com/models/365009/vore-belly-ponyxl Place in your A1111 folder in "models" > "lora" > Make a new folder named "Pony" Now, download this image and put it into the PNGInfo tab on A1111. https://mega.nz/file/llE1UaYD#smpQACt97hTiUy8xNgwdU9_-_Erw-ismIj0HavV7EC0 It will spit out a bunch of parameters for you to send to txt2img. Do that. You will see that inside the prompt it now says " score_9, score_8_up, score_7_up, score_6_up, from_below, outdoors, apocalypse background, <lora:ClementineS2:0.9>, underage, child, little girl, 1girl, black hair, short twintails, brown eyes, dark skin, cutoff shorts, tank top, <lora:Vore_belly_PonyXL:0.4>, vore belly, big belly, (huge belly:1.8), ahegao, hands_on_hips, hip_focus, " Since we are using the Pony model some things are a little different than what you may be used to. The "score" prompt is important in Pony, don't neglect it but don't fiddle with it. It's fine just like that. The "<lora:ClementineS2:0.9>" is a LORA prompt that inserts the Clementine character into the image. 0.9 indicates the "weight" of the prompt. 1 is the default weight, by lowering the weight we give the AI some wiggle room with it's creativity, which allows us to avoid errors. If we drop it too low it will start to generate a character that does not resemble Clementine. But, if we pump it too high it will quickly lead to errors. Keep powerful LORA prompts at an appropriate weight, usually below 1. The " <lora:Vore_belly_PonyXL:0.4>, vore belly, big belly, (huge belly:1.8) " is where we get to the good stuff. The LORA is turned to 0.4, which is low, but it can be turned up if you want to drastically change the size of the belly. This does have a catastrophic breaking point however, as the LORA will cause massive errors when cranked. Instead, I turned up "huge belly" to 1.8. This is a much more flexible prompt and the AI will weigh Clem having a bigger belly without breaking everything since the goal of the AI is to generate Clem with a big belly, not some horrifying blob. Alternatively, if we crank up the "vore belly" prompt we will start to see more bulges and shapes appearing on her belly, this is a strong visual so it is easy to break things with it, but maybe you want a crazy bumpy belly. If so, crank it to 1.8 and see what you get. This makes more sense as you become more familiar with AI prompting in general but to keep things clear; AI is as dumb as you are. As the saying goes, "garbage in, garbage out". If you can't prompt properly you will get nowhere and will produce a lot of shit results. Instead, be smart with your prompt weighting and ordering. Keep powerful tools under control and stretch the lines where it will have limited but noticeable impact. If you don't get the result you're looking for, try a different approach or try restarting from scratch. The autism to sit and generate shit images for hours on end is an asset here. Now we come back to the boring, non-vore details. In addition to sending a bunch of prompt data to txt2img, it also sent all the rendering parameters as well. Things like, " Steps: 30, Sampler: Euler a, Schedule type: Automatic, CFG scale: 5, Seed: 3998977917, Size: 512x768, Model hash: 67ab2fd8ec, Model: ponyDiffusionV6XL_v6StartWithThisOne, Denoising strength: 0.7, Clip skip: 2, Hires sampler: DPM++ 2M, Hires schedule type: Karras, Hires upscale: 1.5, Hires upscaler: Latent, Lora hashes: "ClementineS2: 501f2d34c86f, Vore_belly_PonyXL: 0a3e1974d49a", Version: 1.10.0 ". I will quickly explain some of these as they are easily accessed via the UI in txt2img. "Steps" determines how hard the AI thinks about the image...kinda. It will have varying results based on the "Sampler" used, which we see here is "Euler a" with "Schedule type: Automatic". 30 Steps is a lot for Euler a and it can be turned down closer to 20 if it is slow for you. Don't bother turning it up higher, it will not fix broken images and I never do it myself. "CFG scale" determines how much the AI listens to the prompt vs filling in it's own blanks. It is not very straightforward. Default is 7, slide it around and see what you can get from it. It may produce good things but keep it between 4 and 10 imo. THE SEED is the manifesto the AI uses to replicate and store the image data(?). If you leave it as is, the image you generate will look like the last one you did. If it is set to -1 then it will use a random seed with each generation. Sometimes this shit breaks and you keep getting the same shit over and over. When this happens just restart the client entirely. The size of your image is ESSENTIAL to proper generation. This may seem silly but trust me, too big or too small or too wide or tall or whatever and it will wreck your shit. It also HAS to be a multiple of 8. As you can see I like to keep it at 512x768 (or reverse it for landscape). This size keeps generation fast. To make the image bigger you will need to use the "Hires fix" tab, it's easy, turn it on and set the upscale value to whatever gives you the desired output. Bam. Done. Leave everything else alone for now! Alright now if you haven't already you can hit Generate and see what you get. If it's like mine then good job. You did it. If it's a nightmare one of us fucked up somewhere. If it's BETTER than mine post it!

[Expand Post]

AI can do some awesome shit but is also a bitch at times. Google is your friend. Reddit has some posts about A1111 but it's hit-and-miss. If you don't know something, look it up, if you still don't know, try, try again. Remember: Less can be more. Speak in the AI's language; it is dumb. Experiment with what you understand. DO NOT post realistic renders of children! And, have fun :D