"Quan" is a artist who has created many interesting and adorable chibi-style character portraits for games like Genshin Impact and Arknights. Back when AI-generated art first emerged, I tried training a model to replicate this art style, but the results were less than ideal. Now, with the rapid development of high-quality models, I’ve finally managed to achieve a version that I’m fairly satisfied with.

Since the primary training materials for this model were mostly ainime avatar, with only a few other works included, and my focus was specifically on creating portraits, the model might not perform very well in complex scenes (though I haven’t tested this yet). Additionally, because most of the original images had watermarks and I didn’t do extra processing (at least not for the first version), some outputs from the model may still feature watermarks. However, this doesn’t significantly affect the overall visual quality.

Here are some recommended parameters:

For those who want to use the model to create portraits(like the example shown below):

steps:24-30

CFG:4-7

Lora strength:0.8-1

sampler:eular A

size(when creating portrait):1024x1024 , or at least a square

prompt:

score 9,score 8 up,score 7 up,anime,quan,1 girl,anime avatar,solo,white pupil,white background,{Describe the name and appearance of the character you want to draw.}

This is just a basic example of a prompt to help you quickly and easily get results after acquiring the model. Feel free to explore other ways to use it!

As for negative prompts, I usually don’t use them, but using them might lead to even better results.

That's all for this version. I hope everyone enjoys it! If you have any suggestions or would like to share your work, feel free to leave a comment below.

描述:

训练词语: