This is the second half of a special dialogue between AIST Solutions, part of the AIST Group (National Institute of Advanced Industrial Science and Technology), and Visual Bank Inc. It follows the earlier installment, “Shared Values in Working Behind the Scenes: Collaborating Towards the World’s First Rights-Safe Image Generation AI Model*”

As of July 2024, Amana Images Inc., a company within the Visual Bank Group, and AIST Solutions Inc., a group company of the National Institute of Advanced Industrial Science and Technology (AIST), signed a joint research agreement to develop a Japan-made image generation AI model that minimizes rights risk.

(See the press release here (available in Japanese only)

In the first part of the dialogue, Seiji Osaka, Representative Director of AIST Solutions Inc., and Yoshiro Tanaka, Director-General of the Information and Human Factors Domain at AIST, joined by Ippei Mochizuki, Representative Director of Amana Images Inc. (Visual Bank Group), and Shoma Akamatsu, Tech Lead at Visual Bank, to talk about the background and goals of the project.

In this second dialogue, explores the risks tied to image generation and the participants’ outlook for the future.

Eliminating Rights Risks from AI Applications Across Industries

― Q. “Minimizing rights risk” has been a recurring theme but, what does that risk actually look like in the world of image generation?

【Mr. Mochizuki】
Since generative AI emerged in 2022, growing attention has been placed on how to handle rights such as copyright and portrait rights in the AI field. We are seeing cases where copyrighted works are used for training without permission, and generated images that imitate specific IPs are spreading online. The rules governing these issues are still not fully established.

In this environment, questions like “Are the rights relationships clearly defined?” and “How can we minimize rights risk?” have become key themes for both those developing AI and those using it.

【Mr. Tanaka】
Technological progress in AI is evolving at an incredible pace, so we must constantly adapt to new developments.

【Mr. Mochizuki】
When people talk about “rights,” they often focus only on the legal side, but it is just as important to think more broadly about how creators and the public perceive these issues. It is similar to the concept of compliance. Even if you explain that something is legally permissible for AI training, that does not automatically earn social trust or acceptance.

The challenge of minimizing rights risk in the AI field is fundamental, but it can be difficult for profit-driven companies to dedicate long-term resources to such research. That is why it is so meaningful to pursue this as a continuous, long-term project together with the AIST Group, a public research institution.

【Mr. Tanaka】
As a national research institute, AIST’s mission is to contribute to industry through fundamental research that private companies often cannot take on. That is why we believe it is vital for Japan to unite its strengths in both technology and marketing, moving forward together as one.

【Mr. Mochizuki】
This joint research can help remove rights-related concerns and risks from the use of image-generation AI across industries. Once those fears and uncertainties are lifted, we believe new businesses and ideas will begin to emerge.

Unique Use Cases Made Possible by a Japan-Born Image Generation AI

― this project carries special meaning as a Japan-made image generation AI model. What do you think makes the approach unique?

【Mr. Osaka】
At AIST Solutions, we see Japan as one large company. Our goal is to create something here that will surprise the world. 

Japan already has many forms of high-quality content, such as manga and music. By combining AIST’s role as the technical engine with Visual Bank’s expertise in data, we can build a healthy ecosystem that brings those uniquely Japanese assets to life.

【Mr. Mochizuki】
That is a very inspiring view. If we can make the greatest distinction between our Japan-made image generation AI model and those developed overseas the ability to learn not only Japan’s culture and atmosphere but also its unspoken customs and subtle nuances, we may be able to create uniquely Japanese use cases that no other model can replicate.

【Mr. Akamatsu】
This approach could also play a major role in supporting Japan’s world-leading IP industries, such as manga, anime, and games. When creating manga or animation, the ‘Japanese essence’ or the emotional tone that feels authentically Japanese is an essential cultural element. By developing a model that can express this quality, we hope to build an environment where animators and manga artists can unleash even greater creativity.

【Mr. Osaka】
Just as Steve Jobs was once fascinated by ukiyo-e woodblock prints, Japan has had a sense of beauty that has amazed the world for more than four hundred years. 

Considering that history, it is hard to imagine AI surpassing Japan’s aesthetic sensibility. No matter how technology evolves, human creativity should always stand at the top of the ecosystem.

What the AIST Group’s science and technology and Visual Bank’s assets create together should be tools that accelerate human creativity, not replace it. We must never forget that creators are at the center of this evolution.

― Finally, could you share your outlook for the future?

【Mr. Mochizuki】
To reach our ultimate goal, continued partnerships with research institutions and companies will be essential. To deliver the outcomes of this joint research to a wider range of industries, we plan to collect specialized datasets tailored to each field.

If we can develop foundation models like this one, where rights and ethics are properly safeguarded, we can create a variety of new use cases and expand into diverse business areas. This dialogue has strengthened our determination to become a starting point for innovation.

【Mr. Osaka】
To take this research even further, we need the support of others as well. We want people who share our vision to join us in building an ecosystem that supports a wide range of industries. By inviting like-minded partners to take part, we hope to work together to drive the development of Japan’s industries forward.

Location: WeWork Hibiya Park Front

* Refers to an “image generation AI model that minimizes rights risk by combining fractal-based synthetic images with real-world images.”