Skip to content

In AI Film, the Trailer Comes Before the Feature

Eleven works in the commercial film field of the International AI Creator Award 2026, and four of them are trailers or teasers rather than finished films. The one that began as a one-minute trailer and ended up in Japanese cinemas shows why that order keeps working.

AI-assisted draft. Reviewed and edited by the Phosphene team before publication.

On this page
Blank index cards fanned across a dark walnut desk beside a rolled paper treatment and a dented steel film canister, lit by one hard window light
Blank cards, a rolled treatment, and a can with nothing in it yet.

The International AI Creator Award 2026 sorts its entries into ten fields. The one called commercial film is the field where the work is meant to reach a paying audience rather than a jury, and it drew thirteen applications, the eighth-highest count of the ten. Eleven works made it into the public gallery, against more than 300 across the whole award. Those counts are the organisers' snapshot from 10 November, 02:30 Japan time.

The name is worth reading carefully. In Japanese, 商用映像 means work made for commercial release, not advertising, so the entries are features, series and shorts rather than brand spots. Five of the eleven came from outside Japan: two from Korea, two from Canada, one from Spain.

Then count what is actually in the gallery. Two finished features that reached Japanese cinemas. One trailer for a ten-minute short that screened in a 2025 theatrical omnibus. Two teasers for series that do not exist yet. One teaser for a feature still in production. Four shorts. And one magazine the organisers entered themselves, disclosed as an operator entry.

Four of eleven, then, are trailers or teasers rather than finished films. That is the shape of the professional lane right now: proof before release, and the proof is the submission.

I read the sibling open field a few days ago and came away thinking the mechanism was the work. This field makes a related claim from the other direction. It says the trailer is the work, until somebody commits to the rest.

The trailer is the document that opens the door

The clearest case is Matilda, directed and written by KUMIKO ENDO with editing by Atsutomo Nakagawa, made between Spain and Japan.

It began as a one-minute trailer. That trailer is what started the conversation about releasing it as a feature: the pair then built the film in Barcelona over a few months, generating picture, performance and voice with AI, and it reached Japanese theatres in December 2025. What opened that door was not a script or a deck. It was sixty seconds that showed the film already existing at the quality the release would need.

Anyone who has pitched creative work knows the usual order: write the treatment, get a maybe, get a small amount, prove it, ask for more. The trailer-first route collapses two steps. You spend the budget of a proof on the proof, and the proof becomes the sales argument. It only works if the proof is genuinely at the target quality, which is the part that costs.

All-AI features arrive with the IP already attached

The other finished feature is 怪奇!死肉の男, Tsuyoshi Sone's adaptation of Hideshi Hino's 1986 horror manga, made as a feature entirely with generative AI. ChatGPT wrote the script. The voices are AI, with the lead voiced by Hiroshi Tamura, who starred in Hino's Guinea Pig 2. It opened in Japanese theatres on 24 July 2026 and has been screening around the country since. The secretariat entered it as a recommendation and said so in the write-up.

Note what made that possible. A forty-year-old manga with a cult readership, a director who knows the genre, and a performance reference the audience can recognize. The generative part is the production method, not the reason anyone bought a ticket.

The honest lesson: full generation is viable where the demand already exists and the tooling only removes cost. It is not yet a way to manufacture demand.

The remaining theatrical entry runs the other way on scale. グランマレビト, by Hiroki Yamaguchi at Gaumapix, is a ten-minute film inside generAIdoscope, an omnibus of three AI-made shorts that ran at Uplink Kichijoji from 29 August 2025. The submitted piece is that short's trailer. Ten minutes still reached a screen with tickets attached, which is more than most of this field can claim.

The project that looks institutionally funded is a hybrid

The Taste of Water, from RYO NAKAJIMA at Supersub, is an animated documentary about sake: a manga artist follows his father's line that good sake tastes like water, guided by a yeast spirit. Live action, drawing, 3DCG and generative AI in one pipeline. It is an international co-production with Toonz Media Group, and the entry is a teaser released mid-production.

That mix is not an accident of taste. Co-productions, broadcasters and funds have delivery specs, and a pipeline that can hand over plates, artwork and clean audio in known formats survives a technical review. Pure generation is still awkward to paper. So the projects that reach institutional money tend to be hybrids, and the AI sits where it is cheapest rather than everywhere.

What the solo makers did instead

Four shorts in the list, three of them credited to individual makers, all four using a constraint as the design.

Stone Memory, by Matsuri Yumeai, is a psychological short with no dialogue and almost no human figures. A white stone carries fragments of a life: a garden, a birthday, the sound of water. It was built with GPT Image 2.0, Seedance 2.0, MiniMax Audio and Suno. Removing faces and speech removes the two things AI video is worst at, and the film's subject absorbs the abstraction instead of fighting it.

Junho Her, working alone in Korea, entered two films about the same bond between a person and a Sapsali dog. The second, The Gift, a 12-minute final cut, started from a single photograph posted in his comments by a follower in Mexico. That is roughly the opposite of an IP adaptation, and it still produced a finished film with an audience.

SPEEDHUNTER, by Yves Dalbiez, takes the one thing a Tokyo production can shoot that nobody else can: the Shuto Expressway at night, in fog, with a driver outrunning his own past. It was added to the field on 8 November, the newest entry in it.

None of these needed permission. That is the trade. Constraint-driven shorts buy freedom and reach, and they do not buy a distribution deal.

The numbers that should keep you calm

Eleven works in the gallery had collected eight comments between them at that same snapshot. Two entries drew two comments each, and five drew none.

That is a tiny sample of audience interest, and it is the honest state of the field. A gallery listing is not distribution, the summaries were written by the organisers rather than the makers, and three entries are teasers for something that does not exist yet. Treat the list as a map of what people are attempting, not as proof that any of it worked commercially.

Two entries are finished features with cinema dates attached: Matilda and 怪奇!死肉の男. Everything else is a teaser, a short, or a magazine. If you want a reading of the field's health, that ratio is the reading.

If you want to be in this field next year

Build the sixty-second proof of the longer thing, and make the proof good enough to be the pitch. Choose the version of the project that a real cast, a real location or a pre-sold story already makes credible. Keep the pipeline hybrid enough that a funder's technical review does not end the conversation, and keep one constraint in the short that removes your model's weakest capability rather than working around it on every shot.

Public voting in the field closes 13 November, Anywhere on Earth, with a watch party on AICU's channel at 20:00 Japan time on 13 November. Watching all eleven in a row is a faster education than reading about them, and the gallery is open to anyone.

Sources

Share this guide
X LinkedIn