The true value of AI lies not in being cheap or fast, but in "making the impossible possible."

The fifth session of "dentsu prototyping hub," a workshop for the Dentsu Group hosted by Dentsu Inc.'s creative technologists, focused on "Generative AI."

We held a four-part workshop series featuring Professor Akihiko Shirai from Digital Hollywood University Graduate School / AICU Inc. as the instructor, focusing on learning advanced image generation.

Workshop overview here.
 

We look back on the workshop with Professor Shirai and discuss the future and potential of generative AI. (dentsu prototyping hub, Ryunosuke Ono)

<Table of Contents>
▼ Not reliant on luck or intuition. "Creative AI" infused with human intent.

▼ Why we adopted "Creating creators" as our new mission.

▼ What does generative AI bring to advertising creativity?

▼ Building systems that make creators and rights holders happy.

---

Don't rely on luck or intuition. "Creative AI" infused with human intent.

Onuki: After the four-day workshop, please share your honest impressions.

Shirai: Two things: "We did it!" and "That's amazing!" I refer to the kind of creative work using generative AI we implemented this time as "Creative AI." Utilizing the cloud and remote work to simultaneously teach Creative AI to 175 professionals was probably a task of unprecedented difficulty, even globally.

I teach at university and give various lectures, so I'm used to speaking to large groups. But this was my first time actually getting people of varying skill levels to roll up their sleeves and do creative work.

Axe: This time, we set the difficulty level beyond just "let's try out generative AI a bit" and aimed for proper advanced-level proficiency. Despite strict participation requirements, like needing to register for a paid service, we were amazed by the number of participants, which far exceeded our expectations.

Shirai: It was truly great that most participants stuck with it until the end and created something. The finished works were all fascinating. We packed a lot into those four days, and I think it's amazing that everyone kept up with that pace.

Of course, we also meticulously designed the support system. Our teaching assistants were on standby via chat, carefully guiding participants through tool usage.

Axe : At Dentsu Inc ., we've seen increasing use of generative AI for prototyping—whether proposing concepts to clients or sharing planners' ideas within teams. However, many people don't know how to take it "to the next level," like fine-tuning outputs to match their vision precisely. So it was great that we could create a workshop where no one was left behind .

Shirai: This time, we consciously emphasized "not relying on luck or intuition." We referred to it as Creative AI rather than Generative AI, because if creators can't direct it according to their intent, it becomes an endless task.

That's why we emphasized the so-called "Image to Image" technique this time. In general image generation, "Text to Image" is used, where you input words like "A man and woman toasting with beer." If we consider that as a one-dimensional level, "Image to Image" – where you give instructions using images to generate images – is a two-dimensional level. It allows for more precise instructions on layout, character poses, clothing colors, fingers, and so on.

Axe: Originally, many people used prompts (text instructions) to generate images as a prototyping method. This time, we could challenge creating images from images and refining details to achieve the desired picture, which was very well-received. Since the tool is paid, I thought participation might be a high barrier, but everyone was eager. The overwhelming reaction was, "Just paying this fee lets us do something this fun?!"

Shirai: While we couldn't cover video generation this time, I believe there's demand for video ad prototyping. For instance, it could enable swapping out characters or changing clothing within the same scene. If we hold another workshop, I'd like to explore these areas.

The Reason Behind Adopting "Creating Creators" as Our New Mission

Axe: Professor Shirai, you founded AICU Inc. in the US under the motto "Creating creators." This workshop seems part of that effort. How did this concept originate?

Shirai: I originally developed game engines and conducted research and development in graphics. Later, my focus shifted to VR and the metaverse. However, I felt a sense of stagnation—that even if I created many things I found enjoyable, it didn't significantly change the world. That's why I became a university professor.

At that point, I sensed the potential in a meta-level shift in thinking: moving from "Let's create things I find interesting" to "Let's create people who create things they find interesting." I've been pursuing this activity for the past ten years.

At Digital Hollywood University Graduate School, I run a lab called the Creative AI Lab. Every member there shares this spirit of "creating creators," and each has written a technical book.

Ono: Dentsu Inc. creators are also "makers," but you essentially installed everything for them—from cutting-edge creative techniques to the mindset itself.

Shirai: Exactly. Advertising is a quintessential creative field, but creativity is much broader—everyone is a creator. We're entering an era where those previously called "consumers" become "creators."

For example, we're active in community programming classes, painting workshops, senior programs, and even the " Tomoiki Metaverse" creative workshop for people with disabilities, with Kanagawa Prefecture as our client. Eventually, I believe that as more individuals create things using generative AI – like when someone says, "My grandpa's been really busy lately, and it turns out he's making pictures on the computer!" or "Thanks to that, he seems full of life every day!" – the perception of generative AI itself will become more positive.

But if we lose sight of the joy of human creation, AI—which should be a tool for people—could lead us to live incredibly dull lives. That's why I always focus on creating "makers" who add value that generative AI alone can't produce.

If my activities increase the number of "creators," then even if those people pay 3,000 yen a month for ChatGPT's paid service, services generating value worth more than 3,000 yen will proliferate. Creating things is incredibly fun and stimulates the brain. If all Japanese people can unleash their creative talents, I believe 100 million people could generate value equivalent to 1 billion people.

Axe: I see! I totally get the meaning of "creating creators" now! Also, your perspective on generative AI from a creator's viewpoint was really important.

Shirai: Exactly. With generative AI, it's not just about showcasing its amazing technical capabilities; there are various challenges for widespread adoption. In that sense, I really appreciate Dentsu Inc.'s approach to generative AI.

Axe: Thank you! At Dentsu Inc., we have a dedicated governance organization for generative AI usage. We're preparing a very thorough safety net, carefully considering both legal and ethical perspectives.

What does generative AI bring to advertising creative?

Shirai: Axe, from the perspective of advertising creative, what significance do you think this workshop holds?

Axe: Traditional prototyping centered around one art director using Photoshop to combine existing free assets. By leveraging generative AI, tasks like slightly altering expressions or changing a subject's position—essentially testing variations—can now be done by anyone, not just art directors. That's a key point.

Also, previously, the workflow involved everyone outputting a comp based on what the creator imagined in their head. Going forward, using generative AI to output dozens or hundreds of idea patterns for verification will allow direction like, "This is interesting, let's adjust it this way," while confirming. This is also fascinating. It's a way of creating that involves repeated trial and error from an early stage to get closer to a better output.

Shirai: Generative AI excels at outputting massive variations under identical conditions—it can generate 100 or 200 outputs with the click of a button. Comparing it to traditional creative work, it's like taking many photos and selecting from them. I think this approach is highly suited for advertising creative.

Axe: On the other hand, I don't believe the communication skills cultivated by Dentsu Inc. over time, or the understanding of what moments truly move people's hearts, can be replaced by AI. I think the human abilities to "select" and "direct" will actually accelerate in the era of generative AI.

Shirai: I agree. There's a common misconception that "generative AI reduces effort, so ads will become cheaper." Frankly, that's incorrect. I explicitly state this in my book. The process of selecting the best output from a large volume and then refining it through multiple revisions requires human passion and expertise above all else. We use generative AI precisely because it quickly generates numerous ideas, enabling us to elevate the quality of the final output.

That's precisely why people who originally drew manga or illustrations can master generative AI faster and utilize it more effectively.

Axe: You also gave a thorough explanation on "how we should perceive and engage with generative AI."

Shirai: Regarding the risks of generative AI, I emphasized that it's not a simple yes-or-no decision. Instead, it requires legal and business judgment to identify and resolve issues. To clarify: we must properly organize and analyze factors like laws, business practices, and most importantly, customer sentiment – not just mix them together. Creators shouldn't just rush into production with a "this is great!" attitude; they must respect any anxieties or concerns.

We also discussed how generative AI creativity shouldn't rely on luck or intuition. Simply repeating what's often called "spinning the wheel" will never lead to a conclusion, nor can it be called creative. I explained methods for extracting parameters from generated images for adjustment, and for conducting experiments and testing hypotheses.

Axe: That's right. Dentsu Inc. creators have a strong interest in technology, and I've received a lot of feedback saying it was fun, interesting, and they want to learn more.

Shirai: Everyone was genuinely enthusiastic, weren't they? If you view generative AI merely as something "cheap and easy," the moment you realize "it can do this," you might lose interest, thinking "just getting a decent image is enough." But the Dentsu Inc. team's eyes lit up from that point on. They had an attitude of "I want to use this for my own expression" and "I want to master it to create even more interesting things."

On the final day's hackathon, one piece I awarded featured an image of someone playing guitar. Playing guitar is one of the most challenging things to generate in image creation. Seeing them struggle and try to make it happen, I was genuinely moved, thinking, "Ah, this is a creator." If we were using the "Creator Certification" framework, I'd want to give them an "Advanced Creator Certification" (laughs).

Creating systems that make creators and rights holders happy

Axe: Beyond Dentsu Inc., what are the key points for society as it implements generative AI?

Shirai: While a flood of generative AI services are emerging globally, they haven't yet translated into major business opportunities. However, I believe the key lies in building things incrementally—not just consuming generative AI as a technology, but linking it to intellectual property rights, for example.

Axe: Combining IP protection with generative AI seems to hold promising potential.

Shirai: People often think generative AI is complicated, but I believe we need to explain it in a way even children can understand. For instance, if someone asks, "Can you generate an image of this popular character?" technically, it can be done. But we must always ask: Could this cause harm to someone? Who bears the responsibility?

What we're working on at AICU Inc. right now is precisely building that kind of system. For instance, we should be able to generate images of characters kids love, charge 100 yen per use, and have half of that—50 yen—returned to the rights holder. We might be able to create a system where creators get proper profits and everyone is satisfied.

Conversely, pirate-like or bandit-like actions that take things without any respect are obviously wrong. For example, using a character in videos that violate public order and morals, for political advocacy, or for criminal activities – I believe the right way to use technology is when rights holders can control these "things that shouldn't be done or things we don't want done."

Axe: I agree. We need to proceed cautiously and build something truly positive. Generative AI still has many unresolved areas and challenges. Rather than promoting it unreservedly, we must carefully examine and address each aspect when implementing anything. That's why Dentsu Inc. has established an internal consultation window for AI-related matters from legal and ethical perspectives and distributed generative AI guidelines. What are your thoughts on such initiatives?

Shirai: I think that's excellent. At AICU Inc., we consult with companies worldwide, and a common scenario is that only the executives or top creators—those driving new initiatives—dive headfirst into new technologies, leaving the majority of employees behind.

First, it's essential to share what the top people are doing with everyone, clearly show the company's governance and direction, and then create an environment where people can get hands-on and learn. In that regard, Dentsu Inc. feels like it's not just at the level of trying to adopt the latest trendy technology; I sense a high level of understanding of generative AI at the organizational level.

Axe: Thank you! I believe it's especially important to have a strong stance on thoroughly verifying that creators' rights are not infringed upon. I hope it becomes widely known that Dentsu Inc. has a dedicated governance organization for generative AI and is approaching it with caution.

Shirai: At Dentsu Inc., you're prioritizing legal and compliance matters while still proactively engaging with generative AI technology, right? Plus, there's a willingness to listen to concerns about generative AI. I used the word "respect" in the workshop to explain this—it's crucial to proceed while acknowledging opposing views as valuable, rather than just saying, "Well, the tech can do it, so why not just do it?" In that sense, Dentsu Inc.'s creative team members, many of whom studied at art universities, understand the value and meaning of human craftsmanship. I think this whole approach is built on that respect for their expertise.

Actually, as an evolution of these workshops, I'd like to eventually create something like a "Professional Certification Exam for Image Generation AI." It would be a licensing system at the practitioner certification level, guaranteeing technical proficiency and compliance. Achieving this would allow licensed professionals to work properly, while conversely, those engaging in pirate-like activities would be visible as "operating without a license."

Axe: That would certainly make companies feel more secure about adopting generative AI. Also, no matter how much generative AI evolves, it doesn't mean we can just rely on it for convenience. Human creativity and the ability to move hearts and minds come first. I was reminded that generative AI exists to help us leverage those human qualities even more effectively. Thank you so much for today!

*Mr. Shirai's opinions do not represent the official stance of Dentsu Inc.

---

Generative AI School Overview

・DAY1 "AI Utilization at Dentsu Inc. / Sharing Use Cases"

A workshop was held to share Dentsu Inc.'s AI strategy, AI governance, and risk case studies. This elevated participants' AI literacy before they began hands-on work with generative AI.
&nbsp;

・DAY2 "Generative AI Fundamentals: Text-Based Generative AI Basics"

Participants learned the fundamentals of image generation AI, including how to write prompts correctly and utilize negative prompts.
&nbsp;

・DAY3 "Advanced Generative AI: Creating Images as Intended"

Learned techniques like "creating images from images" and "controlling poses and expressions." Practiced methods for controlling tone and manner, adjusting impressions, atmosphere, and emotions associated with brands, products, or services—down to facial expressions and finger gestures.
&nbsp;

・DAY4 "Generative AI Practice: Hackathon"

As an online hackathon, a comprehensive exercise was conducted within the limited timeframe of two hours. Participants reviewed functions while controlling elements like model selection, identical outfits, the same person, and different situations to complete the key visuals they wanted to generate.

&nbsp;