Package Design Proposal with Generative AI Reduced from Two Weeks to 15 Seconds! — A Traditional Packaging Manufacturer’s Challenge in Developing Generative AI Features [Iwasaki Shiki]

Mr. Michieda:
We are a comprehensive packaging manufacturer located in Hasami Town, Nagasaki Prefecture. We have two types of packaging: simple folding boxes and hand-wrapped boxes, but we are particularly focused on producing high-quality, original hand-wrapped boxes. Our packaging is often used by restaurants, confectionery shops, and for packing ceramics such as Hasami ware and Arita ware found in our local area.

――Could you tell us about the background of developing the AI-based package design generation system?

Mr. Michieda:
Our company, established 64 years ago, started with the processing of cardboard boxes and gradually increased the quality of packaging to enhance its value. In the packaging industry, it's common to leave the design to design companies. However, we pride ourselves on our high level of planning ability and have focused on design creation ourselves.

However, we faced issues with the time and labor involved in the design process. In the traditional production flow, after meeting with the client, the sales team would convey the details to the designer, who would then work on the design. It took about two weeks from the meeting to propose a design, and an additional week for revisions.

Although we are relatively fast compared to other packaging manufacturers, there were still cases where clients would say it took too long. As a result, we experienced a lot of missed opportunities due to the time spent on design proposals.

Also, sometimes clients would change their mind during the two weeks before the proposal. There were cases where clients canceled after the revisions because the final image was different from what they expected. It was a significant burden for designers to create multiple design proposals without knowing if they would lead to a contract, only for the designs to undergo repeated revisions.

Solving Our Own Challenges with Generative AI

――Could you tell us what sparked the start of the project?

Mr. Michieda:
The trigger was our participation in the “JSB New Business Creation Program 2023” organized by Shinnwa Bank. It is a 9-month program where core regional companies and startups in Nagasaki work on open innovation. During this program, we thought about how to solve our own challenges.

In that process, we thought about how we could solve the design-related challenges. We wanted to make the matching of package images smoother to reduce the burden on designers and allow sales teams to make proposals on the spot.

――Why did you focus on generative AI?

Mr. Michieda:
Our CEO came across a news story where a major beverage manufacturer used generative AI to design labels, and he brought the idea to us, asking if we could use it in our company. When I actually tried a generative AI app, it quickly generated dozens of design options in just seconds, and I thought we could use this for package design as well.

However, before starting development, we researched generative AI ourselves, but the technology was still in development, and there wasn’t much information available, making it difficult to judge what was correct. So, we had concerns about tackling this project alone.

An Unexpected Presentation in the First Meeting: “Did They Really Think This Through?”

――I understand that your company had no prior experience in system development. Were there any concerns about that?

Mr. Michieda:
To be honest, I was filled with concerns (laughs). However, Shinnwa Bank, which hosted the project, and the supporting companies provided extensive support. Also, Solashi genuinely considered our company’s needs, making proposals and development with great care, so I felt comfortable proceeding with them.

――Why did you choose Solashi as your development partner?

Mr. Michieda:
Initially, the project’s support company selected a list of startups that excelled in generative AI. From that list, we narrowed it down to eight companies and met with them, and Solashi was one of them.

The first impression of Solashi was one of surprise. During the first meeting, Mr. Shimazoe immediately gave a presentation. He thoroughly considered everything, from what we were thinking to what could happen next, and even demonstrated the concept on the spot. While other companies were still testing the waters, I was shocked that Solashi had thought so thoroughly and came so prepared.

――What was the decisive factor in your decision?

Mr. Michieda:
First and foremost, it was Mr. Shimazoe's personality. During the interview, I could feel that he truly cared about our company. He was easy to talk to, and when I shared our thoughts, he immediately predicted the future several steps ahead and made suggestions. He even came all the way to Fukuoka, and I was impressed by his enthusiasm.

Utilizing Existing Technology for a Practical Approach to Functionality

――What choice did you make regarding the development method?

Mr. Michieda:
In the end, we decided to add functions specialized for package design to existing generative AI services. I had received proposals from other companies to build generative AI from scratch, but the budget was quite high. Additionally, we realized that there was no significant difference in quality compared to utilizing existing AI technology. Considering cost performance, we felt it was more reasonable to use existing technology.

Later on, during the development, we initially planned to use Stable Diffusion, but based on a suggestion from Solashi, we switched to a system utilizing the more effective Dall-E 3. After making the request, we mostly left everything to Solashi, and they proceeded without any issues.

――Were there any difficulties during the development process?

Mr. Michieda:
Tuning the shape of the generated packages was difficult. For example, when I instructed to generate a square box, the box that came out at first was completely different from what I had in mind.

So, I had Solashi teach me how to better instruct the AI, and through the AI's own learning, they were able to produce results that matched my expectations.

They also made many adjustments to the system to support the Japanese language, improving its ease of use.

A Design Proposal That Took Two Weeks Now Takes 15 Seconds

――Can you tell us about the changes since the system was implemented?

Mr. Michieda:
Now, during meetings with clients, we can generate designs on the spot while listening to their requests. The design proposals come up in about 15 seconds, so not only can we propose them on the spot, but we can also come up with alternative suggestions while observing the client's reactions.

Being able to generate designs in front of the clients has made the proposals more concrete, and I truly feel that it has become a very powerful sales tool.

――How has the feedback been from the designers?

Mr. Michieda:
Previously, a lot of time and effort went into image proposals, leaving limited time for the actual design work. However, by using this function, we can reduce the man-hours spent on proposals, allowing us to focus more on the actual production. Designers have evaluated this as enabling us to produce higher-quality designs within the limited time available.

――Were there any unexpected benefits?

Mr. Michieda:
During a logo design project, when I showed the client a design generated by AI on the spot, the agency praised it by saying, "The range of design options has expanded and it's helpful." I believe we were able to overturn the preconceived notion that package manufacturers lack design capabilities.

Not Only for In-House Use, But Also a New Business as a Service

――What are your thoughts on the future expansion of the service?

Mr. Michieda:
Currently, we are mainly making proposals to our existing clients, but in the future, we want to establish a system that can handle requests from a wider range of clients.

Additionally, not only for in-house use, but we are also working on a business plan for external sales of the function. Currently, it is in the verification stage, and we are working on concrete plans with the goal of releasing the service in the spring of 2025.

――Which companies are you targeting with the service?

Mr. Michieda:
Initially, we are targeting other package manufacturers in the same industry. Many package manufacturers face design challenges and have given up on solving them. By allowing other manufacturers to use this design generation service, we hope it will serve as a catalyst for revitalizing the entire industry.

――What is the most important factor when choosing a system development partner?

Mr. Michieda:
I feel that the most important factor is "the ability to propose with foresight." It's not just about technical skills, but about being able to show how to utilize that technology and present a clear vision of the future.

Also, flexibility in thinking is important. In our project, we realized that by utilizing existing systems, we could provide a practical solution without building a system from scratch. I greatly appreciate that we were presented with such a cost-effective option.

Additionally, trust is an essential element. We had no experience in system development and were quite uncertain. In that context, Solashi was very supportive, communicated carefully, and made us feel at ease. From that attitude, we naturally built trust, and it became clear we could rely on them.

To maintain a long-lasting relationship, I realized once again how important compatibility and trust between people are.