/vt/ - Virtual Youtubers

>>36751714
I'm on the WebUI with NAI leaked version

Prompt:

masterpiece, best_quality, nsfw, (1girl), (rratatat-NAI-t16-10000:0.95), red hair, messy hair, (multicolored hair:1.1), (streaked hair:1.2), white hair, (twintails:1.2), sharp teeth, open mouth, mouse ears, tail, nude, breasts, spread legs, (sex), vaginal, penis, (testicles), pussy, pussy lips, fat mons, clitoris, female pubic hair, pussy juice
Negative prompt: hair ornament, chain, puffy nipples, collar, lowres, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, out of frame, extra fingers, mutated hands, (poorly drawn hands), (poorly drawn face), (mutation), (deformed breasts), (ugly), blurry, (bad anatomy), (bad proportions), (extra limbs), cloned face, censored
Steps: 30, Sampler: DPM++ 2M, CFG scale: 9, Seed: 1920052602, Size: 1024x1536, Model hash: 925997e9, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, First pass size: 512x768

Used embeddings: Rratatat-NAI-t16-10000 [667d]

>>36752024
I'm using some of the features on the Automatic1111 Stable Diffusion Web UI to help finetune my generations. If you are on the Web UI, got to your Settings tab and in the "Upscaling" section make sure the "Upscale latent space image when doing hires. fix" option is enabled, and save your changes.

Here is a short summary of my process:

- Start with a prompt to get the general scenario you have in mind, here I was just looking to seggs the rrat so I used the embed here >>36743515 and described some of her character features to help steer the AI (in this case hair details, sharp teeth, her mouse ears and tail) as well as making her be naked and having vaginal sex
- Generate images at a default resolution size (512 by X pixels) at a relative standard number of steps (30 in this case) and keep going until I find an image thats in a position I like (in this case seed 1920052602 gave me a very nice one to work with, as you can see here https://files.catbox.moe/8z2mua.png))
- Copy the seed of the image and paste it into the Seed field on the Web UI, which will maintain the composition of the image. I then double the resolution I was working with (so here I went from 512 by 768 to 1024 by 1536) and checkmark the "Hires fix option" underneath the width and height sliders. Hires fix is the secret sauce on the Web UI that helps maintain the detail of the image when you are upscaling the resolution of the image, and combined with that Upscale latent space option I mentioned earlier it really enhances the detail. With that done you can generate the upscaled image.
- Play around with the weights of the prompt tags and add things to the negatives to fix little things like hair being too red, tummy too chubby, etc. You have to be careful with adding new tags because that can drastically change the image