Prompt: bakemonogatari_style, studio_shaft, anime_coloring, shiny skin, hachikuji mayoi, standing on one leg, pinup pose, cat hands pose, cleft of venus, innie pussy, small breasts, nipples, nude, hips, thighs, loli, cute, black hair, hair ties, twintails, smile, detailed, best quality, sexy, extremely detailed, official art, white background, cowboy shot, side view,, Negative Prompt: 3d, bad anatomy, bad hands, (blurry), lips, text, border, jpeg artifacts, saturated, extra fingers, watermark, pubes, unfinished, large breasts, backpack, censored, 1boy, male, headband, Model: waiNSFWIllustrious_v90, Seed: 649908228, Steps: 30, CFG Scale: 5, Aspect Ratio: 9:16 (Tall), Width: 768, Height: 1344, Sampler: Euler Ancestral (Randomizing), Scheduler: Normal, Refiner Model: waiNSFWIllustrious_v90, Refiner Control Percentage: 0.2, Refiner Steps: 60, Refiner CFG Scale: 5.5, Refiner Method: Post-Apply (Normal), Refiner Upscale: 2, Refiner Upscale Method: Model: RealESRGAN_x4plus_anime_6B.pth, Automatic VAE: true, LoRAs: 0: Bakemonogatari-multiple_characters-PDXL, 1: watanabe.pony,, LoRA Weights: 0: 0.8, 1: 0.4
I took the output and made some edits manually with clip studio. Edited the eyes, hands, hair, toes, and nipples. I like to upscale during generation; seems to work best for me with a low control percentage and high steps.
All I can really say is to download the same model (tPonynai3) & Style lora (Vixon's Pony Styles - petra) and copy the metadata (Specifically: Prompt, Steps, Sampler, Schedule Type, CFG Scale) then it should generate a similar image. I use SD Webui Forge which was pretty easy to install, just make sure you have a decent graphics card.
All I can really say is to download the same model (tPonynai3) & Style lora (Vixon's Pony Styles - petra) and copy the metadata (Specifically: Prompt, Steps, Sampler, Schedule Type, CFG Scale) then it should generate a similar image. I use SD Webui Forge which was pretty easy to install, just make sure you have a decent graphics card.
I am not sure what aibooru picks for metadata, but the model used is wrong. it's an unholy cross of naturalnoob and kurimomi. probably fetched the model fromthe sd prompt saver which I never bother to assign it into.
UPD: yeah, it definitely is picking stuff from sd prompt saver. it can be entirely disregarded as I never fill anything but the seed and folder to save to
From looking at your other images, I assume you used img2img, and some image editor as well? This is pretty nice.
Ah yeah, I just use Krita for merging images. The key things I recommend are: comfyui-impact-pack's maskdetailer, comfyui-prompt-control, and merging your own models w/ ModelMergeSimple (part of Comfy-Core) to get the art style that you really want.
i can hardly express how much i look forward to discovering new posts by you whenever i visit here, youre easily my most favourite creator. thank you, thank you!
Sorry,I don‘t know how to teach,I’m not good at English.But you can try combining different artist prompts or LORA to create your own unique art style.（AI Translation）
What ai do you use? Can you teach me how to use promts?
It's just that in my opinion, Loli is too tall, I would like to make her shorter.
Hi! I didn't notice it, but now that I look at it I think you're right. Personally, I would have fixed it with Photoshop if I'd noticed it before posting.
You're fortunate that your tummy has stayed slim; not everyone can achieve that when they gain weight. That said, maybe lay off the doughnuts a little to keep it that way.
Separate LLM that rasterizes latent image. Although given that you're using service instead of proper UI, I'd assume you have no control over such parameters and instead have very high CFG scale, poor weight scaling, or bad LoRA or checkpoint outright. In other words - skill issue.
A VAE isn't a LLM, it's an encoder. It's in the name.
@VaporWaifu99 Why is the watermark so big? It's bigger than the post description on the bottom right.
Wanted it to be big enough that you can't miss it, but see through enough that it doesn't totally take over the image. The text in the bottom right is inspired by TCGs, it's a set and card number. I feel like those tend to be small on cards...
@Archive That's right, and I think human-made art will always remain of more value than generated, even if the technical quality becomes identical. Maybe gens will be mainstream unlike today, but it'll still be considered cheap mass-produced stuff, kind of like fastfood compared to culinary masterpieces. That said, whichever happens, it doesn't have to stop us from developing better models.
@Archive Shitty hentai OVAs are not the "quality threshold" to strive for, and are not the only kind of hentai out there. In training and assessment of generative models, we should look up to animations done by enthusiastic artists (which is exactly what iodoff compared it to), not by underdog japanese studios looking just to sell their another batch of low-effort DVDs on next comiket
Yeah, but that's is a pretty dystopian future that I rather not think and talk about. It will destroy the existing business model, talent/source material pipelines of the anime industry. I am comfortable with AI replacing hentai OVA studios, but Ghibli? No. KyoAni? No. Trigger? No. Etc. I still want media that I am we are not jerking off to to be hand drawn.
@Archive Shitty hentai OVAs are not the "quality threshold" to strive for, and are not the only kind of hentai out there. In training and assessment of generative models, we should look up to animations done by enthusiastic artists (which is exactly what iodoff compared it to), not by underdog japanese studios looking just to sell their another batch of low-effort DVDs on next comiket
imho still looks like shit compared to real animation
It just need to reach the threshold of "good enough to replace the vast majority of animators who pumps out shitty hentai OVA slop", and it is close to it, probably faster that teaching actually a human novice to animate those hentai slops.
Actual anime, compared to hentai, is a different story. But Toei, aka one piece' studio, is already looking into using AI on "Various Processes of Animation Production".
