Girls' Frontline FLUX Edition _ Girls' Frontline Character Fan Models

This time using the characters of "Girls' Frontline" as the training dataset.

Explore the parameters of the training.

FLUX actually has many complex parameter settings,

Each training session can only research one type,

and because there is no comparison between multiple trainings of the same content,

we can only rely on reasoning to speculate on the purpose of the parameters.

And because when using it, FLUX's positive cue phrases are divided into two boxes.

t5xx box and clip box, their functions are actually quite different.

The t5xx box is mainly used for stylistic descriptions that affect the entire image,

while the clip box is for specific content descriptions.

You can also use a set of cue phrases with both boxes simultaneously.

The effects need to be gradually explored through extensive testing.

Other than hardware thresholds, there are also other usage thresholds.

The Niu Mao will share some user experience every time it is uploaded.

Originally, I didn't like using trigger words for training, but now I am slowly getting used to it.

Everyone can first refer to the given parameter specifications and try to generate images.

If you can use the full-blooded DEV model for image generation, the effect will be better.

To generate images faster, I used the FP8 version, sacrificing some quality.