Model Introduction

This is a model designed for all your anatomy-heavy inpainting jobs (especially NSFW oriented). As there aren't many specially trained inpainting models on human bodies and anatomy, this one will provide more detailed and higher quality results than any conversion of a T2I checkpoint every will. The training utilizes masked training which has shown to give better results faster since it removes unnecessary details from the training images.

How to use the model

As this is an inpainting model, you of course have to use it that way. But more specifically, I'd recommend to use "inpaint only masked" function in your UI of choice (I use Forge) with a resolution around 1024x1024 (1280x1280 can give very high quality results, but not always as coherent with the original image). The best aspect ratio can vary depending on the image, but I usually do a 8:5 one (1024x1600).
Denoising strength: 1

Masked content: original

CFG: 3 (usually the best)

Sampling method: DPM++ 2M SDE

Masked padding: 64 (change depending on the resolution of the image, increase for more context in complex scenes, but decreases detail)

Sampling steps: 20-30