messed around with illustrious a bit, super fast to run and theres some decent LORAs on civit ai already, definitely reccomend for a non-realistic look

>>43997 You might want to use the standard SDXL VAE, you have lots of white spots in your images, and it could also be why the results are very grey and dull.

>>43999 Not 'op' but curious. Is that a base Illustrious VAE issue? I haven't seen that using whatever VAE red dot pops out of the node.

>>43999 im using comfyui with and have literally never touched the VAE node, not even sure how to lol

>>43946 It was trained on base Illustrious XL 0.1 model. I know about this problem and I need to fix the dataset. This is mainly caused if there is no "close-up" tag in the prompt

Any advice for keeping background characters out of diapers when I add them into prompts? I tend to create images that have 1 or 2 characters focused with people in the background, but I want to be able to more freely control which character is wearing a diaper. Still new to this and having some trouble.

The new Kling Image to video model works pretty well although it's only a few seconds of video on the free tier. This was from a flux.dev image I made and kling did the rest.

Best i can do with a phone

>>44053 FINALLY the rule of 3D is over, fucking about time.

Have any of you tested NoobAI (especially v-pred 1.0) and know how to set it up to make it work well (sampler, scheduler, cfg rescale) or even know what settings to give LoRA training?

>>44186 Training on the EPS one gives worse results than training on base Illustrious. Results are the same quality on both NAI-XL branches so no need to adjust anything for lora training. The recommended settings on Civit provide fine results. The only thing that seems a little different is that results tend to be better when prompting an artist name as well. Easiest way is to look at some of the example image prompts and see what artists people are using. You don't need "artist:" even though plenty of people use it. Just the name is fine. Some names with styles that I liked are "cato monocatienus, re-ghotion, fumio rsqkr, merunyaa, zd_1995, OlchaS, sakimichan, massakasama, bigdead, and aetherion". I wrote a random prompt generating script, rather chatgpt did, that uses danbooru tags that I used for sample images. You can see the comfyui workflows I used along with the settings on Civit. Here's the 4 for the attached images. masterpiece, best quality, newest, 1girl, (diaper:1.2), aetherion, star-symbol-background, hair-between-eyes, heart, grin, standing, long-hair, crop-top-overhang, makeup, long-sleeves, open-clothes, brown-eyes, feather-coat, looking-at-viewer, short-hair, brown-hair, hair-ornament, hand-on-own-cheek, cleavage, green-eyes, japanese-clothes masterpiece, best quality, newest, 1girl, (diaper:1.2), bigdead, blue-hair, standing, long-hair, sky, black-thighhighs, tail, gloves, makeup, long-hair, cow-ears, hair-between-eyes, wind, smile, sleeves-rolled-up, puffy-long-sleeves, choker, side-ponytail masterpiece, best quality, newest, 1girl, furry, (diaper:1.2), merunyaa, smile, armpits, large-breasts, miniskirt, pumps, jacket, sky, feet, smile, breasts, corruption, looking-at-viewer, skirt, long-hair, toes, black-jacket, brown-shirt, animal-hood, sweater, short-hair, purple-eyes, anthro masterpiece, best quality, newest, 1girl, (diaper:1.2), facing-away, (fumio rsqkr:0.75), blazer, short-hair, small-breasts, spotlight, glasses, flower, ponytail, arms-at-sides, hair-spread-out

>>44189 >>44186 Totally botched selecting the right images for the prompts... Here are the other images and the other prompts. You should be able to figure out which is which. masterpiece, best quality, newest, absurdres, highres, 1girl, (diaper:1.2), (cato monocatienus:0.9), re-ghotion, facing viewer, full-body, yellow shirt, breasts, green hat, short hair, blonde hair, wavy hair, grin, hands behind back, outside, sunny day masterpiece, best quality, newest, 1girl, (diaper:1.2), zd_1995, OlchaS, sakimichan, sakimichan \(style\), sleeveless, gradient-hair, blue-eyes, red-hair, purple-hair, shirt, two-tone-hair, tail, frown, gloves, solo, bra-visible-through-clothes, blush, nipples, multicolored-clothes, striped-clothes masterpiece, best quality, newest, 1girl, (diaper:1.2), (massakasama:0.4), off-shoulder, chinese-clothes, bdsm, collarbone, striped-clothes, outdoors, purple-eyes, solo, navel, pink-background, black-hair, thighs, jewelry, head-tilt, short-sleeves, blue-hair

>>44107 >>44106 omg these are top notch

>>44364 that looks awesome, what did you use to create it? I would add some hypno prompts to make it even better

trying out some flux images thrown into Kling

Tried to run flux on RTX 4060 8GB but it's too slow. should I wait for the 5x cards? prompt: score_9, score_8_up, score_7_up, woman in a diaper, pacifier, nipples Negative prompt: low-res, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation Steps: 20, Sampler: Euler a, Schedule type: Automatic, CFG scale: 7, Seed: 2316800683, Size: 1000x1200, Model hash: 8cd86b11ad, Model: ponyDiffusionV6XL_v6StartWithThisOne, Clip skip: 2, Version: f2.0.1v1.10.1-previous-635-gf5330788

>>44370 Krea.AI. I like because you can use img2img and realtime. I use this style "#znrrzfslpp" polaroid, a shirtless woman with a pink pacifier in her mouth in crib touching disposable heavily plastic diapered white nappy with with toy pattern, a man in a suit touches her face #znrrzfslpp https://www.krea.ai/apps/image/realtime

>>44443 For training you'll need much more VRAM - for Flux at least 16GB VRAM. I tried to do it myself on RTX3070 8GB with Pony and it took 8 hours of training. You can always try to use third party solution like civitai or rent a cloud GPU on runpod.

>>44457 oh nice I'll look into runpod.. .17*10*30*6=306 == used 3060

>>44189 >>44191 I did my own research and these samplers and schedulers do a great job - most of my testing was done on `waiSHUFFLENOOB_vPred01` and `noobaiXLNAIXL_vPred10Version` checkpoints. These nodes are needed as well: ModelSamplingDiscrete (with ZSNR and v_preditcion enabled) and RescaleCFG. I also noticed that if prompt is short or the enviroment/background is not well defined, then the output images became too vivid or there are weird shapes popping out. The solution for this is to expand prompt, better describe background or rise the value of RescaleCFG by 0.1 or more

>>44443 For generating flux does take longer than Ponydiff, but with a smaller card the output reso makes a big diff for both. If you like flux try lowering the resolution to the recommended portrait/landscape and it may go a chunk faster. Naturally also depends on whatever else you're loading up, for an 8GB card will be hard for FluxDev

What is the best site or program to use, in your opinion, to generate more images out of a given image prompt?

used this prompt young woman, black hear, big breasts, no shirt, wearing a puffy diaper, starting at a mirror. A big crib in the room, lots of toys on the ground, the reflection in the mirror is a herself becoming a lot younger, sucking her thumb all failed attempts so dont bother much in unspoilering this shit is hard though, also generating video takes ages and consumes a lot of buzz

>>44627 It'd help if your English was better, anon

inpaint goodness

🐎

ntrMIXIllustriousXL_v40 is pretty good, when you know what you're doing

>>44646 why are you horny for Jeets, anon?

bump limit?