The model to diffuse them all.

Just a bayesian merge of the best models around (chosen by me).

Very stable. Very diffusion.

Listens well to the prompt. Doesn’t need too much of the prompt.

Can do anything, but mostly pussies and women, so kind of NSFW.

Resolution

Up to 960×1200, higher – you’re starting to get double face/torso/anything

(+ after a brief test with Jennifer Lawrence, starting to lose the likeness,

but no one stops you from experimenting)

Samplers

Blurry, but stable:

Euler or Euler a (more creative)

Sharp, but things can go wrong:

DPM++ 2M Karras or DPM++ 2M SDE Karras (more creative, needs more steps)

Settings

Enable quantization in K samplers for sharper and cleaner results

Eta noise seed delta: 31337 (doesn’t do anything, just cool number)

VAE

Baked VAE, wasn’t converted to fp16, can be used with –no-half-vae

Recommended

Extensions:

Self Attention Guidance (patch is not needed since webui version 1.3.0)

Booru tag autocompletion

3D Openpose Editor (for ControlNet)

Environment variables:

PYTORCH_CUDA_ALLOC_CONF="backend:cudaMallocAsync"
CUDA_MODULE_LOADING="LAZY"

LoRAs:

Detail Tweaker

Saturation Tweaker

Sharpness Tweaker

Brightness Tweaker

Example images

All examples are made on a dev branch of AUTOMATIC1111’s webui

with –xformers –xformers-flash-attention

and Random number generator source set to CPU

“The circle test”

Introduced by the guy on reddit

Have fun and don’t do harm with it.

And while you’re at it please post your generations, so that other people could try them.