Japan’s first fairly-trained certified AI image generation model is released, with the concept of a cute VTuber at art school 

Abstract Engine inc. has released Japan’s first fairly trained-certified AI image generation model. Called Mitsua Likes, its figurehead is the company’s AI VTuber Elan Mitsua. Under the theme is “Art made with everyone,” the concept is that users will ethically provide images for art school student Elan Mitsua to learn from, thus influencing what the AI model produces. It has been officially certified by the US non-profit Fairly Trained (as reported by Otaku Souken).  

Mitsua Likes does not use any existing AI models as its foundation and is trained only on images that have been explicitly authorized or are verified public domain. It is not trained on AI-generated data from other AI learning models either. To ensure transparency, all the training data is made public on the official Discord. Furthermore, images generated by the model have clear credits identifying the source image authors.  

So far, Mitsua Likes has been trained on over 58,000 images. From January 2025, the company will start accepting submissions of new images for the AI to learn from. You can find out more about Mitsua Likes on the official English language website.