What an AI Film Festival Is Actually For

The industry session at AI Film Festival Japan 2026 started right after Phil Tippett's live segment, which is a hard act to follow. Two speakers took the stage at Tokyo Innovation Base on 2 November for the festival's third day: Kenichi Watanabe, who works in what he calls value design, and Akihiko Shirai, the AICU CEO who has spent about 30 years across games research, teaching and publishing. The host was Matsui from the festival.

The conversation was supposed to be about where humans fit in AI entertainment. It turned into something more useful: an argument about what a finished video is worth once anyone can make one.

Information arrives. Meaning does not.

Watanabe's opening claim was the sharpest thing said in the hour. Information can be delivered, he said. Meaning cannot, because meaning is generated inside each person. When technology runs ahead of that, what arrives is only information: a slightly prettier image, slightly better music. The work of design is building the conditions where something can emerge in the viewer, not declaring what they should feel. Telling an audience "you will be moved by this" might work as information. It will never produce the feeling.

That sounds abstract until you apply it to a prompt. A clip that announces its own emotion is asking the viewer to receive a message. A clip that withholds something is asking them to complete it. The second one is why a two-shot of two people not talking can hold attention longer than a technically superior render of a city. Nothing about the render changed. The amount of work left for the viewer changed.

The catalog argument

Shirai's answer to the same question was practical: keep a record.

His organisation, AICU, publishes a monthly magazine of roughly 150 pages, plus festival catalogs. The reason is not nostalgia for print. On NicoNico, YouTube and X, work flows past and disappears. You cannot later reconstruct what was made in which month, with which model, by whom, and why it mattered. Human memory will not do it. So the catalog does.

He offered a concrete example. At AIFJ 2025, a 70-minute feature called Matilda: The Devil's Gene, made by two people, was screened and then released theatrically in December 2025. That is eleven months ago. Without a date and a source attached to it, the fact becomes hard to find, then hard to believe, then gone. Two people made a feature and the record of it is already thin.

This is the least glamorous advice in AI video and probably the most durable. Put the date on the work. Put your name on it. Keep a page that lists what you made and what you used. If you have ever tried to reconstruct your own process from six months ago, you already know why. The pipeline notes are not paperwork. They are the only version of the work that survives the feed.

What 400 entries looked like

Shirai was reviewing around 400 submissions for the International AI Creator Award while preparing the session, and he named what stood out: ordinary life.

His example was a video by a newlywed woman about how much she enjoys grocery shopping. Not a spectacle, not a franchise, not a fantasy epic. A small domestic pleasure that nobody had ever paid to put on film, because there was no market for it. Now there is, at least for the person who made it and the people who recognise themselves in it.

That is a bigger shift than it sounds. For decades the cost of production decided whose inner life got recorded. A supermarket run was not worth a camera crew, so it was not worth representing. When the marginal cost of a scene approaches the cost of deciding to make it, the set of lives that get documented changes. Shirai's framing is that consumers are turning into creative consumers, and that the people who spent one to five years making a game or a film now have to work alongside them.

Volume, and the bot problem

The award data also shows what happens when submission is cheap. One creator entered 10 works across five categories. Another entered eight across three. In a normal film festival, you submit your best single piece; here, prolific creators curate their own catalog and submit the catalog. Volume is the new normal: by mid-October the award had passed 303 entries from 171 creators, which is where our earlier read of the data started.

That has a side effect festival organisers should plan for. The award runs a public vote, and Shirai reported that someone attacked it with roughly 300 bots. He mentions it without drama, as a design constraint: any open vote in this field will be gamed, so the format has to survive it.

For a creator, the lesson cuts both ways. Volume gets you into the room, but self-curation is the actual skill. Ten mediocre entries place you in ten categories and win none. The people who show up with a body of work have usually spent as much time deciding what not to submit.

The icebergs are melting, the culture is not

Shirai's metaphor for the last two years is global warming. GPUs produce heat, the heat becomes the vibe, and what melts is the iceberg of existing industry. Video production is melting. Publishing is melting: a monthly 150-page magazine now comes out of a company of two or three people.

He borrows a line from media artist Kazuhiko Hachiya, later quoted by Yoichi Ochiai: media art melted into culture. His addition is the second half, which he repeated as the session's title. Culture does not melt. The sentence is not reassurance that solid things stay solid. Read it as a demand on anyone working in a medium: if your form dissolved into the general culture, what is left that is specifically yours?

Watanabe's version of the same idea comes from Japanese performance. He argues that the core skill of Japanese entertainment is transcription, the act of becoming something else: masks in kagura, animals in the Chōjū-jinbutsu-giga scrolls, the whole machinery of reincarnation stories. The technology changes what transcription costs. It does not change why people want it.

Anti-AI is a room, not a wall

The most grounded part of the hour was about resistance inside institutions. Shirai teaches, and he describes a familiar scene: educators who state plainly that AI is evil, students using ChatGPT to test compositions and colors while hiding it from the people around them.

His approach is not argument. It is work. When he teaches veterans who built games from the original PlayStation era, the ones who said they would never accept this, the conversion happens when they see a tool that can hold their drawing. The sentence he quotes from them is about capability, not ethics: with this technology and an assistant that can draw, maybe I could direct this alone.

That is a better internal pitch than any policy document. Show a professional the thing they could not do before, in their own style, and let them decide. Watanabe added the structural reason resistance is loud: knowledge is tied to authority, status and income, so any technology that distributes knowledge will be resisted hardest by whoever held it. He points at tea ceremony as a case of a tradition that survived by inventing a new protocol (the first kettle ceremony of the year) rather than by pretending nothing had changed.

What to take into the next project

Four things, none of which require a better model.

Write the meaning gap into the piece. Decide what the viewer has to bring, and leave room for them to bring it. A shot that explains itself is information.

Date and attribute everything. A public index of your work, with tools and dates, is cheap now and impossible to reconstruct later. It is also the first thing a client's legal team will ask for.

Curate before you submit. Cheap production makes volume easy and selection hard. The catalog is the portfolio.

Assume the vote, the comment section and the leaderboard can be gamed. Build the format so that a bot rush does not decide the outcome, and build your own expectations the same way.

Shirai closed by saying that at the end of 2026 the basic skill is coding, and that the people who will be fine are the ones who build an experience instead of generating a clip and waiting. That is a strong claim from someone who runs an AI film festival. It is also consistent with everything else he said: the render is the cheap part now, and the part that behaves like a memory is the part people keep.