Generating images from prompts can be a complex task, and achieving high-quality results without proper assistance (such as our configurations) can be quite challenging. Particularly, when dealing with ambiguous prompts like 'a girl,' 'anime,' and so on.
After thousands of trials and certain amount of research, we have come to a conclusion that prompts structured as follows are more likely to yield good images:
{adjective} {emotion} {Subject}, {behavior}, {background}
By abstracting much of the complexity and handling the heavy lifting for you, you can solely focus on transforming your imagination into real artwork.
Ideally, your prompt should still include at least two components:
For example:
'A girl with pink hair'
For ambiguous prompts (without: pre-configurations, style, color, behaviour, posture, etc.)
We have provided some of the pre-config styles that have been tested and deemed as stable (relatively), please do at least pick one, if you want to have more freedom on that, do feel free to include your desired styles in the textfield. In that case, you may want to choose the 'default' style in the Style section to avoid unwanted 'crossover'.
When saying specific, background seems to be something people have often left behind. Background as one of the most important element of an image, it is strongly recommended to include at least some details, such as:
"An small inn locate at the foot of the mountain, bright shining sun hanging up on the sky"
It really helps when you include emotions into the generation, AI are sometimes bad at drawing faces and turns out to be everything looks awesome, except the face.
Good example would be like
a middle age asian man, sadly standing in the middle of the street (Style - 90s)
Long story short