Fun with AI - Special tricks in Image Creation: Prompt Attention and Classifier-Free Guidance
Made by the Stable Diffusion AI

Fun with AI - Special tricks in Image Creation: Prompt Attention and Classifier-Free Guidance

Lets see the effects of parameters in AI image creation.

The prompt is

A scene from a Scifi movie. A space ship with blazing guns

We are using the Interface Stable Diffusion and the model technoRealism_v10.

The pictures were rendered in 512x512.

The graphics is NIVIDA A4000 with 16 GB VRam on a HP z440 workstation


The standard output looks like


Seed: 3529848616

or


Seed: 3529848617

New Prompt attention

Lets use this prompt

A scene from a Scifi movie. A (space) ship with blazing guns

We have here put the term space in parantesis.

This lets the model pay more attention to the term space

The result is


Seed: 520969607

or


Seed: 520969609

More prompt attention

We use as prompt

A scene from a Scifi movie. A (space:1.5) ship with blazing guns

In this prompt, we put a number as indicator, how much the model should pay attention to space.

Seed: 1689003750

or


Seed: 1689003749


Modify the Classifier-Free Guidance scale (CFG)

This time, we have the original prompt and change the cfg scale

Until now, we had the cfg scale to 3.5


But now, lets change it to 10.3

This will change the output quite a lot.

The result looks like


Seed: 2062818308

or


Seed: 2062818306


cfg = 49.9?

If you change cfg to the max of 49.9 (in this interface)

the output looks like


Seed: 140716866

Not very impressive!

要查看或添加评论,请登录

Reiner Merz的更多文章

社区洞察