>>47800 'Diaper Styles - Tiny Diapers' on civitai for flux.

Hopefully I'll get the photos done for the merries that are currently on the model this weekend. Unfortunately, baby diaper patterns don't have names. Anyone got a better name than "Merries Neighborhood?"

what kind of model and prompt are you people using?

>>47930 BRO IM SO EXCITED FOR THIS

sometimes it comes out neat sometimes it makes you want to claw your eyes out

>>48125 Holo the Ditzy Pup

any more bedwetting ai art ? feels like we barely get any sleeping content.

I made some chubby gay ABDL art.

>>48195 .....WTF is this shit? Please, for the love of God, don't bring up this faggot shit again. Seriously, who the hell finds this shit sexy?!

>>48201 >who the hell finds this shit sexy Gay guys I asssume

>>48201 It's garbage, but the fact it's like 40 fucking images of the same thing is the real problem.

>>48201 I'm not attracted to men, but I'm not angry about other people being attracted to men. This is weird but given the nature of our fetish it's bound to happen.

>>48220 thats literally everything with ai though

>>45317 > Illustrious pleaseeeee share what main checkpoint model and lora checkpoint and prompts you are using for these!!! <3

>>48290 i'll tolerate dudes in diapers if they're somewhat attractive. these are all skinnyfat nerd virgins. actually, a hard cock bulge in a diaper is hot, but just make the guys have abs and be in shape and not look so fucking bad.

Am I able to create something with my current setup? 1660 Super I5 11400F 16GB Ram If not, is there some reliable cloud thing like the one from google? I saw someone using it a long time ago but couldn't get how to use it properly, it would be ideal if I could run it for like a hour or two just to create some images, I've been using NovelAI to make some chat bot interactions and I wanted to create some images based on those interactions

>>48372 Probably, just grab webforge UI for flux.dev/illustrious or foocus for anything older than that. They are both one click installers. Your card is old though so it will probably take a few minutes to generate anything on any newer models.

aight whos gonna try the new openai 4o image gen

>>48390 That one part in the Snow White remake where she has been cleaning for hours and her diaper is too full to take anymore from her, then the evil queen arrives to tell her its time for a change and that step-mom knows best, to which Snow White reluctantly agrees, and then is happy that it happened.

>>48290 Not OP, but it's flux and the pullup lora on civit. Prompt is likely just a description of what is shown. "a girl on a bed wearing a pullup" or something like that. >>48399 It would reject the request. It's not open source or local.

>>48390 What would you recommend? The ones you posted look pretty good although i'd prefer a more anime look, is the difference between webforge and foocus that big? And do I risk my GPU using it too much to generate stuff? I'm comfortable with like 2 to 3 minutes of wait, anything way longer than that would start bothering me Also worth mentioning I would need to use my HDD since my SSD is almost full (bg3 is a bitch)

>>48493 Foocus is easier to use but it can't run the flux or illustrious models so you will be stuck with sdxl or lower models and no one is really making loras for them anymore. You really need to just start using it to figure out what you want. civitai has a ton of diaper loras you can use. TLDR: Just fucking try one of them.

>>48499 Tried ComfyUI since I saw people saying it was the less heavy one, pretty nice but really it takes too long and the UI is a bit weird to deal with, also tried webforge UI with flux.dev but didn't try diapers with it (just a simple cat photo took 15 minutes), ComfyUI takes around 13 to 15 minutes to make one image for me, I used Diaper Styles from Civitai, I'd like to know what LoRa you used for the Pokemon and the Alice ones you did, I liked their style especially the Alice one

>>48509 15 minutes is ridiculous. There are ways of getting flux to run well on very low vram gpus. Google flux and your gpu to find instructions (or ask ChatGPT). Words like Quant, FP8, GGUF means you are on the right track. All the anime ones are going to be illustrious based models which is itself an SDXL based model. It should run much better compared to flux which is an enormous model in comparison. NoobAI and WAI-NSFW are great illustrious fine tunes that I would recommend trying first. They can do diapers and characters normally, but there are plenty of loras available for more control. If you are getting cloth backgrounds from the loras, the weight is too high or the description is too short for flux. Flux likes very wordy prompts.

>>48509 Any good anime AI you are seeing is probably done with illustrious which is just a heavily trained flux. It uses booru tags and there are a few diaper loras for it on civit. With my 8gb 3070 I can do illustrious or flux generation in a minute or two on webforge and sdxl generation in about 30 seconds or less on foocus or webforge. Assuming you don't have a potato PC do some searching for the best GPU driver for image generation. There used to be an old driver you needed to use.

>>48531 Why do you call it flux? Pony, Illustrious and NoobAI are all SDXL trained models.

>>48532 Oops, yeah that is an sdxl model my bad. In any case it's better than pony models IMHO although both can do not bad diapers without a lora.

>>48531 Kind of managed to do it, the faces look horrendous though, had tons of problems with flux over something to do with clips and VAEs, I tried using Afrodite for that but yeah no chance even had a blue screen once These are from waiNSFWIllustrious_v120, it took me a while to figure out Pony prompts my friends used to use aren't the same as Illustrious ones, it took me 3 min to do these the prompts were: masterpiece, best quality, amazing quality, very aesthetic, 1_girl, diaper, teenager, auburn_hair, camping_tent, full_body, small_breasts <lora:Plain_Diapers_Illustrious:1> Negative Prompts:worst quality, comic, multiple views, bad quality, low quality, lowres, displeasing, very displeasing, bad anatomy, bad hands, scan artifacts, monochrome, greyscale, twitter username, jpeg artifacts, 2koma, 4koma, guro, extra digits, fewer digits, jaggy lines, unclear Any chance I can improve on these? The faces especially the eyes still bother me quite a bit and I can't seem to figure out how to do different styles, there were some really good ones on the White Diapers Illustrious homepage on civitai

>>48547 Not exactly sure with WaiNSFW, but the base Illustrious model is trained without using "_" for tags, so adding _ reduces the efficiency of the tag and can even remove it entirely. For styles you can prompt artists directly and you can of course prompt for stuff like watercolour and photorealistic. Different models can also have inherent biases towards certain styles if not prompted otherwise. Also unfortunately anything below 768x768 will have reduced quality as the models have a base training of image resolutions 768 and above.I like to use 832x896 for example. (also I prefer DDPM as sampler and "Align Your Steps GITS" as scheduler, but most choices do well and it's really just RNG when one works really well with a certain prompt and seed and when it doesn't. Just avoid "DPM++ SDE", it has the best quality but the worst speed by far) In many cases not having stuff in the negatives can be better than having a lot, or simply just using 1-3 things. Example negatives to test: 1. [black:white] dress, clothes pile, close up 2. gym, white dress, tucked in shirt, frilled apron, bloomers, bow panties, suspenders, harness, inflatable toy, print, stickers 3. bedroom, bed, 3d model, hanging panties, diaper mattress, napkin, body pillow, pillow, lunchbox, (frame:1.4), toilet, square, rectangle 3,5. bedroom, 3d model, hanging panties I wouldn't use "teenager" either, at best "teen" but if it isn't specifically loli/shota or mature/milf then just describe the body attributes you're looking for without mentioning age-range as it's not trained on those. You can then put unwanted body/age results in the negatives. Boobs too big? Put big breasts in negatives. Body too young? Put loli/toddler/shortstack in negatives, etc etc

>>48547 Here you can see some examples of how volatile AI can be and how there is never really a definitive answer to anything. In the first image you can see how wildly it can differ when changing up the prompt in relatively minor ways. In the second you can see how much the Lora itself can influence and "overtake" the image in quality, style and composition. Thirdly you can see how it can change with different negatives and a resolution change. And what works well for one model might not work as well for another model despite having the same base.

>>48552 Actually pretty good now all changes imrpoved it a lot, still not able to do a bunch of styles though just these more generic ones and the faces and diaper prints are kinda weird if I look up close but definitely better The changes increased the time a bit though, I was doing the images in 3 minutes before and now they take around 5 to 7 minutes with increased resolution and different sampler and scheduler I think the real problem was the Plain Diapers Lora though, it was messing up the art even after the changes at 1 strength, first two pictures are from real-ish diaper 2.0 with 1 strength and 3rd and 4th are from Plain Diapers but with 0.5 strength and some of the negatives you recommended Still not able to do stuff like the alice one from above post nor the one from plain diapers homepage though ( 5th and last image)

>>48562 Also, managed to get flux to "work" now instead of blue screen or error message I get 40 minute generations for a blank or completely blurry image So yeah, don't know how to make that one work yet

>>48562 Yes naturally it's a big increase in resolution so it will take longer unfortunately. What interface are you using? Forge (and ReForge) are based around being optimized and fast, but there is no fixing slow hardware in the end. https://github.com/lllyasviel/stable-diffusion-webui-forge If you mean Alice from >>48499 that's a lot of PieceOfSoap's style, whilst the facial features has more of a... I can't remember the name but I''ve seen that smirky mouth and pointy noise combo somewhere Piecofsoap is not really trained on in Illustrious, but the plain diapers lora is clearly trained on pieceofsoap, so you would just have to experiment around or hope that guy tells you. There are some illustrious models with a more western bias to their default style too.

>>48562 To get that Alice pic I used the plain diapers lora plus an art style in my prompt to modify it. Maybe disney? I forget since I just do a ton of batches with a ton of different styles. Illustrious is trained on so much drawn art you can get wildly different results just by changing artists in the prompt. There are plugins for webforge that help with tags as well. The one I use is called webui-tagcomplete I think. https://mzmaxam.github.io/Illustruous-Artists/index.html shows a bunch of different artists with the same prompt.