Cut through AI noise with weekly insights from Japan to AI-proof your career or business. We aim to bring you insights that are: digestible, actionable, and unique (breaking through the information firewall in Japan).

In this week’s newsletter:

For the past couple of years, AI creators have been quite obsessed with prompts. Longer prompts, meta prompts, prompt engineering, prompt libraries. But experts now add a further attribute to their creation, especially with AI video: references.

Prompt can be easily created, shared, saved, and bought as prompt packs. Most big AI Youtubers will give you their prompts in the comments for the walkthrough videos. You can even buy prompts from prompt marketplaces like promptbase.com or create them with tools like custom GPTs.

But experts are increasingly using examples to get better, more consistent and more controllable outputs. These are called references, and AI video is becoming reference-first.

So what are the signals to pay attention to here?

A prompt like: “cinematic, handheld, moody lighting”
can still leaves the model guessing about things like:

A reference image or clip reduces or removes that ambiguity. Instead of interpreting your taste, the model replicates concrete creative decisions.

Words explain what you want, but references show the model what you mean.

Modern video models are not like human directors, they are more like pattern-matchers trained on massive amounts of data (in this case, visual data).

This is why giving an example clip and prompting something like, “Make it like this, but slower” works pretty well, compared to just prompting say “Slow, emotional, atmospheric pacing” which can give all kinds of results.

The models are far more reliable when connected to visual precedent.

Prompt-first workflows have a lot of misses, tweaking, and usage of credits.

Reference-first workflows help anchor the output visually, from where you make smaller adjustments to get to your desired end point quicker.

Particularly for AI video generation, which still burns relatively large numbers of credits, references help make your creation more cost effective.

Anyone trying to create a series, channel or recognizable visual identity may hit the same problem with prompt-only workflows.. they might lose consistency.

References enable consistent framing and a recognizable style. They are anchors. This is why advanced creators are building reference libraries, not just prompt docs.

Prompting is teachable, copyable and increasingly standardized

But referencing requires:

its just harder to be good at that. Two creators can use the same tool and similar prompts, but with different references they’ll end up with dramatically different results.

Taste is going to be the skill that is prized.

AI video is moving from prompt-first to reference-first workflows.
As models improve, words alone are no longer enough to control motion, framing, and style. Creators who rely on references iterate faster, spend fewer credits, and achieve visual consistency that prompt-only workflows struggle to match. Prompt engineering was the firs step, but reference engineering is where taste will elevate creators.

We spoke to Hero Kominato, a community leader in Tokyo who runs one of the biggest creative AI communities, AIMovie.Studio - about his thoughts on how AI is empowering individuals and his mission to empower a new generation of creators

If you've been following AI events and the creative AI space in Tokyo, you may have noticed Hero san’s events, usually monthly, which are often workshop style opportunities for newbies to use AI to create their own video. I ran into him once at a Shibuya Startup Support event last year, and was surprised to find he often runs his events at the Impact HUB, a pretty cool co-working space that I used to work out of as well back in the way. I asked him about how he got started in creative AI, and where he wanted to take his community.

From Freelance Coder to AI Creative: Hero was a freelance software engineer (still is) back in 2023 when MidJourney and ChatGPT came onto the scene. He remembers being captivated by the potential of these tools - the idea that anyone could create stunning visuals or engage in natural language interactions without needing any technical expertise was a key inflection point for him.

As a lifelong film buff, he saw an opportunity to merge his technical know-how with his passion for storytelling and filmmaking. Maybe he could innovate film making with AI, like James Cameron did with visual effects!

Launching AIMovie.studio: A creative community to master storytelling with AI

“Our community goal is to help people to get started with creative AI. So I see many people who have never touched with films or camera, but they come to events and they see the potential of what they can create. I think it's proof that people started adapting to these AI tools. Anyone can start it now, I think."

I asked his thoughts about AI slop - if anyone can create content, what’s the real edge for creating meaningful content with AI?

Hero had some thoughts about predictions of AI progress and what that means for content creation.
“So let's say that we are still in the age that we can distinguish what's real and what's not. I think in that case, people can still make their own content. And my belief is that if you shoot yourself with a camera and use AI for visual effects, so not fully AI generated videos, I think that content becomes very unique.”

“On the other hand, if we reach the age that everything seems fully AI generated, if you can put yourself into the content somehow, I think that will make it more authentic.”

He does mention that, once we cannot distinguish AI from reality, the picture is a bit more unclear. One movement that he thinks will grow is “competitions” where human creators showcase their work. So rather than trying to stand out in an endless scroll of content online, you’ll be more in an engaged creator community showing off your work. Hero is organizing these kinds of competitions going forward as well.

The other burning question I had was about ethics - aren’t human creators going to struggle as AI is used to do their job, and also to “own” their IP?

“I think what we see so far is that, top creators who are adapting to AI are thinking of AI as just “tools”.. and I think if you're talented and if you have creativity in your mind, if you can describe what you imagine is, I think you have an ability to make your content much better with AI.”

And these people, these top creators, I think they understand that AI is just a tool for them. So I think it just doesn't replace them. I don't think AI will replace them in the near future.”

“And as for the ethics, I think it's a very difficult question because, I mean, I also work with actors and so on. So I understand the fear behind, you know, AI taking all the jobs and copying all the work. And I think that fear is definitely there. And even the AI creators worry about if the AI models can just take their work, right? I think when you're making content on their platform, you're basically giving them ideas and your imagination and your data. So, yeah, I think that's a really important discussion. We might need to talk more deeply about it.”

Do you think AI creators can actually monetize their work?

“I think they can, but it’s getting harder. Since everyone wants to be a creator now, it’s a very competitive space and I think there are less people with technical skills who can effectively use coding tools or no-code tools.

On that note, what would you recommend to people getting started in AI creative work?

“I think many creators start like what I mentioned above. I know of some top AI video creators who made projects recently for which they didn't get paid. Even some of the best creators. But yeah, I think you just need to have like one or two good pieces of work, a portfolio that you can show that you're very good at creating these things. And when you have that, think you're ready to work for commissioned projects.”

Hero mentioned that in Japan, there's still some latency compared to other countries, but Japanese people are now started to get more informed about these tools. So the opportunities in this space will start to grow.

Tell us more about AIMovie.Studio and how to get involved!

Hero mentioned they do mainly 2 things:

The Tool: Celsys AI Brush (part of Clip Studio Paint)

What it is: While most of the world is fixated with AI that replaces the artist, Celsys is building AI that assists the hand. Their "AI Brush" engine is not a typical "prompt-to-image" generator. Instead, it’s a suite of surgical AI tools. They can do things like, intelligent "Running Color" blending that mimics how physical ink bleeds into paper, or their "AI Colorization" that can instantly suggest palettes based on your line art. It’s about taking the tedious parts of digital painting (like flatting colors or cleaning up shaky lines) and automating them while artists still keep 100% of the creative control.

Who made it: Developed by Celsys, a Tokyo-based powerhouse founded in 1991. They are major players in the Japanese creative software scene. Their flagship product, Clip Studio Paint, reached a staggering 60 million users globally (!!!) as of January 2026, and becoming an industry standard for manga, webtoons, and 2D animation.

The Philosophy: Celsys took a very "Japanese" stand in the AI wars. In 2022, after community backlash, they pivoted away from generative "image prompts" to focus strictly on "Monozukuri" (the spirit of making things). Their AI is designed to feel like a high-end physical tool.

Why you should try it: If you’re a creator who actually enjoys the act of drawing but hates the "grind" of digital production, this could be a superpower for you. Their latest 2026 updates include "AI-powered perspective rulers" and "smart-fill" tools that can close gaps in your line art automatically. For Celsys, the entry point is actually through their flagship software, Clip Studio Paint. Since they emphasize "AI as a tool" rather than "AI as a generator," the features are baked directly into the brush engine.

An in-person discussion and networking event organized by The AI Collective and btrax in Tokyo. The session focuses on how organizations and teams can thrive in the AI era by rethinking structures, collaboration, and talent strategies.

Featured speakers include global and Japan-based leaders in AI, organizational design, and DX, who will explore topics such as characteristics of AI-ready companies, evolving ways of working with AI, required skill sets, changes in decision-making and evaluation, and approaches for transforming traditional organizations.

The event includes networking time before and after the main session and is aimed at business leaders, HR professionals, and managers seeking insight into building innovation-driven teams with AI at the core.

Details:

The results of the 2026 Common Test for University Admissions (Japan's equivalent of the SAT) were analyzed. While AI scored perfectly in math, it failed spectacularly on Japanese Geography and World Maps.

The reason: Visual hallucination. The AI repeatedly "invented" geographic features and misinterpreted irregular graphical information, proving that even "super-intelligent" models still struggle with the physical reality of Japan’s landscape.

Source: Nikkei - AI takes Japan's University Entrance Exams