In this guide, we looked at how to generate the character we need on a live incarnation, as well as how various result tags are inherent.

As an example, we will use one character to show how it changes throughout the guide. Take the following phrases:

Girl on a white background with black hair:
Girl on a white background with black hair

As a result, we get an image:

We are not completely satisfied with the result, but we like the main part.

We wanted a different eye color, let's make the eyes golden.

Girl on a white background with black hair, golden eyes
Girl on a white background with black hair, gold eyes

But the image now doesn't look like the previous result because the golden eyes had a big effect on the query

We need to create a stable base
we select the main part of the query with parentheses - In this example, the main part is a girl with black hair and a white background, and not an addition to the girl herself:
(Girl on a white background with black hair), gold eyes

The main element is highlighted with additional brackets. In this case, we only want to complement the girl with golden eyes and not remake her.
((Girl on a white background) with black hair), gold eyes

Now we can get variations of this image

We also advise you to reduce the strength of colors, especially if there are several of them, square brackets are used for this - the result will be the same as above, but now it will be easier to work with colors.
(Girl on a [white] background) with [black] hair), [gold] eyes

Next, we can use another interesting thing in the neural network - to use force in character drawing frames. This can for example be used to overlay different colors or tags partially. For this, a construction with square brackets is used inside which there are different tags - blue:gold and the strength of the left side also through “:” 0.1.

Here's how to experiment further: