This is a project started time ago.

This textual inversion is a negative textual inversion for image2image, allow you to port image to realism.

What you need?

1) A good realistic model

2) this Textual Inversion

3) source image

How it work:

Simply use promt: photo of a subject

where subject is the subjec of image(see example).

Negative prompt: kkw-mmr-neg

Settings for i2i:

cfg:30

denoise: 0.52

steps:21

Note: this is setting worked better on my models, on different models you need to try and find your best value.

It also work in t2i as negative, but it try to give realistic output, so if you prompt "oil painting" for example, you don't get oil painting.

Some image, especially flatter image, need 2 or 3 step. so you generate, resend result to i2i and generate again.

Share your result and impression, giving review help me to work it better.