G-STAR's AI Design Project from a CG Designer's Perspective

Hello, this is Nagashima from Omnis. In this blog post, I'd like to discuss an article about the first-ever AI-designed denim couture piece by the denim brand G-STAR RAW, and share the reactions of two of our CG designers to this project.

About G-STAR RAW and G-STAR

G-STAR RAW is a brand that was born in Amsterdam, the Netherlands, in 1989. It established its reputation as a denim brand by releasing the world's first "3D-cut" denim. G-STAR is the company that operates the wholesale business of clothing and goods under the "G-Star RAW" brand. The company is headquartered in Amsterdam, Netherlands, and established a Japan corporation known as "G-Star International Co., Ltd." in 2001.

Introduction to the G-STAR Article

In G-STAR's article titled "G-STAR's First AI-Designed Denim Couture Piece," the journey from exploring designs using AI technology (Midjourney) to the creation of an actual denim cape, as well as the behind-the-scenes processes of production are discussed. They have even shared the prompts used.

Introduction to the G-STAR RAW Video

In this video, you can see how a digital design took shape into a tangible couture piece. The video has four scenes, each representing a different stage of the production process.

• SCENE 1: Designers input information into the AI tool
• SCENE 2: Tailors create the denim garment
• SCENE 3: They work on the finishing details
• SCENE 4: The first AI garment is worn in the studio

We asked our production team members for their thoughts on G-STAR's initiative.

We interviewed two members of our CG team who have knowledge of apparel. Both Muto and Tanami have a background in apparel and have transitioned to become CG designers.

The New Era of Manufacturing Seen in G-STAR x AI

Nagashima: Could you share how you came across the G-STAR article and your impressions after watching the video?

Tanami: I noticed it when it came up on LINE's fashion news. The title with "AI" caught my eye. I was already starting to use Midjourney at the time, so I think I was particularly sensitive to it. I was interested in how AI would design, and I was interested in what would happen when an apparel company incorporated AI, so I continued reading the article. The production process and workflow are presented in a story-like manner, and the way the video is presented is very skillful!

Muto: My first impression was that it's quite challenging to create a silhouette like G-STAR RAW from scratch, and that the unique, unexpected designs generated by AI can look this cool when given form. When I looked at the details of the image generated by Midjourney and the actual garment made, there were differences, but the overall impression wasn't that different and the consistency was maintained. The sewing and fabric quality were high, and the finish was so good that it's not easy to achieve this quality. As a brand specialized in denim, I felt they were very familiar with the characteristics of the material and knew how to make the most of it.

The Difficulty of Making Real Clothes from Midjourney

Nagashima: We usually make clothes in CG, but would you two find it interesting to create an actual product like G-STAR RAW?

Tanami: There are some images that I'd like to try making for real. Like, what kind of fabric would this be, is it even fabric, how is it connected, etc. It would be difficult to make an exact replica of an image created by AI, but I think it would be fun to think about how to do it. Since we can't wear CG, I'd like to make an actual product and try it on.

Nagashima: Muto, you also work as a pattern maker outside, so you could probably make it on your own, right?

Muto: I might not want to try making it all by myself. When you work alone, you lack diverse perspectives, so personally, I think I'd probably find the finished product a bit "lacking". G-STAR RAW was created by a team of professionals, so I don't think I could easily replicate that. It might be easier if you have a production-like environment with designers and other people familiar with the process.

Nagashima: Indeed, G-STAR has a designer team and a Midjourney team in-house. Having such an environment might be crucial. Is it challenging to create a physical product from a design made by AI?

Muto: I think it's challenging. You need to direct the process towards the final product, so considering those elements as well, it's difficult. But, looking at G-STAR RAW, I realized that it's possible to reach this level, and it was good to know that you can design from the original AI image without changing the impression too much and even upgrading it. I'm not pessimistic, but I do wonder what would happen if I were to create it myself.

What's the compatibility between Midjourney and CG production?

Nagashima: When creating CG from a generated image (design) by Midjourney, how does it differ from real-world production?

Muto: It's not easy to compare, but with CG, there can be complications due to simulations and interferences, making it quite difficult to reproduce complex details. Things like draping can't be done with CG. Hmm... that part is a bit challenging... haha. Right now, even simple things like pinning to confirm silhouettes can be difficult with the way CLO is used.

Tanami: Indeed, there are such difficulties. That said, whether it's CG or real, creating something like the G-STAR RAW cape we saw earlier seems challenging in its own ways. In the video, the clothes underneath the cape were also meticulously made, which I found impressive. Seeing that, I thought that if you're only making what's visible in CG, it might actually be easier.

A world of once-in-a-lifetime design created by AI and humans

Nagashima: I tried generating images with Midjourney using the prompts listed in the G-STAR article, but the result didn't look like G-STAR's. As seen in the video, I think they're adding and adjusting the necessary elements to the representative prompts, and I personally find that part fascinating. We even heard the term "prompt artist" within OMNIS. It's interesting how the results change depending on how the prompts are entered and who enters them.

Muto: Whether the item that comes out is good, or the model is good, or the background is good... The impression you get completely changes depending on what comes out. When looking at everyone's generated images from Midjourney, it's interesting to see each person's individuality. The reason why you didn't get a picture that felt like G-STAR might be due to the difference in how they chose from many generated images.

Nagashima: That's true. The method of selection after image generation is also important. What do you think, Tanami?

Tanami: When a design is clearly in my mind, I can just draw it myself. However, I find it interesting to use Midjourney when I can't yet imagine the image. Midjourney assists like a partner or a consultant in design, so there's a sense that anyone can become a designer. On the other hand, without a certain level of knowledge about how to set up Midjourney or about design, you might just end up with a "good image by chance."

Nagashima: That randomness might be considered one of the charms of Midjourney. Its characteristic of never generating the same image twice could be seen as a strength if you look at its rarity. Because anyone can use Midjourney, it tests the user's knowledge and sense.

Creating new things with "AI × human" allows people to do what they couldn't before. As a design tool with potential, it would be great to continue using Midjourney in various ways!

Finally, we each tried generating images with Midjourney using G-STAR's public prompts.

The various designs that came out were interesting!

AI-created Denim Cape

denim, model wear, organic shapes, minimalist, contrasting, advanced and novel armor, techwear fashion, white background, 4K --v 4

denim, model wear, organic shapes, minimalist, contrasting, techwear fashion, white background, 4K --v 5