Download

AI art image generation AIs Emi and Manga Diffusion, which are available for commercial use, have been released for free.

Introduction

※This article is the same as the one published on Note.

https://note.com/aipicasso/n/ncd7c133c3bc2

Hello, we are the development team of AI Picasso. Today, we have an important announcement for you. We are releasing Emi (Ethereal master of illustration), an AI image-generating AI for AI art, for commercial use with no charge. This image generation AI can be used for commercial purposes and has not learned unauthorized images through additional training.
Emi is available free of charge at the following URL:

https://huggingface.co/spaces/aipicasso/emi-latest-demo

Emi can also be downloaded for free at the following URL:

https://huggingface.co/aipicasso/emi

In addition, we are also releasing the concept proof version of Manga Diffusion, a clean image-generation AI, at the same time as Emi. This image-generation AI has learned only from public domain images and images authorized by copyright holders.

https://huggingface.co/aipicasso/manga-diffusion-poc

(The image generating AI is also referred to as a model.)

Features of Emi

1. Specialized for AI Art

This model is specialized for generating AI art such as anime and manga. We have made every effort to generate high-quality images using the state-of-the-art development equipment H100, image generation Stable Diffusion XL 1.0, and AI Picasso's know-how. As a reference, below are some examples of simple generation:

For comparison purposes, the following example shows the same content generated by the previous model and the current model. The new generated image is about twice as high-resolution as the previous one and has been adjusted to the recent art style.

2. No unauthorized images used in additional training

There are models for AI art that use unauthorized, reprinted images from sites such as Danbooru for additional training, as disclosed on X (Twitter). However, Emi does not use unauthorized, reprinted images for additional training.

3. Available for commercial use

Regarding the license, unlike the previous models, it is available for commercial use. There are two reasons for this.
The first reason is that we feel that people are increasingly adopting manners so as not to have a negative impact on the creative industry as image generation AI becomes more popular.
The second reason is that if other image generation AIs are commercially available, there is no practical effectiveness in having an non-commercial license since some models that are commercially available may contain leaked information, which may violate the Anti-Unfair Competition Act. This model does not have any such concerns, and users can use it with peace of mind. Also, to make it easy for users to use this model with peace of mind, we have changed the license to the same general license as Stable Diffusion XL 1.0.

Activities with Creators

Collaboration with Creators

We work closely with creators and have introduced systems for appropriate compensation for providing data and profit sharing for generated illustrations. One such effort is AI irasutoya. We have consulted with Mr. Takashi Mifune, who operates irasutoya, for about a year to provide images and their descriptions for learning. We are working hard to build long-term relationships with creators and earn their trust.

Dialogue with Creators

Emi was created while listening to the voices of creators. When we released the first version, we had active discussions on X. We released the demo first and checked for problems before releasing the AI. We are now discussing the way models, including this model, are developed with creators on X and Discord.

Development of Clean AI

The development team has been dedicated to the development of "clean AI" so that creators do not have to worry about copyright issues or resistance. We have released a test model, Clean Diffusion, which is the world's first image-generation AI created using only public domain images.

The development of Clean Diffusion began with an individual. However, we are now discussing it with multiple companies worldwide involved in the development of image-generation AI and considering it as a social issue.

Development of Manga Diffusion

When aiming for "clean AI", it is necessary to solve copyright issues. Manga diffusion is a model that solves all copyright issues. Manga Diffusion adds the Manga 109s data set, a manga data set that has been approved for learning by Mr. Aizawa of the University of Tokyo and copyright holders, to the base model that learns only from public domain or CC-0 images. However, due to the limitations of expression, this is a concept proof, which aims to demonstrate what is technically feasible. The concept proof version of Manga Diffusion can generate images like the following:

Currently, it is intended to be used for generating backgrounds. We plan to develop more models that are convenient for creators to use, based on this model and CC-0 3D models, in the future.

Summary

Emi and Manga Diffusion are image-generation AIs that are available for commercial use and intended for the future of AI art. There may be pros and cons, but we believe in the potential of these image-generation AIs. If you have any questions or concerns, please contact us at the following contact point: