The Social Network "Y" — Guaranteed to Go Viral

Challenge

“NOT ME THAT WENT VIRAL” is a suspense film about a man hunted by the internet for a crime he knows nothing about. Conventional advertising could not convey the core of the film — the terror of being unreasonably dragged into an online firestorm. Rather than explaining how frightening it is, we needed an approach that made audiences experience it as the target.

Approach

We built “Y,” a fictional social network where an AI auto-generates an unreasonable flame war from a single everyday photo uploaded by the user — a participatory simulation that escalates through three phases.

The experience closes with a single line — “In this world, you can get flamed over something completely baseless.” — and connects straight into the film’s trailer.

Technology

To guarantee both the quality and the safety of the experience, we built the system on the following stack.

AI Engine (Gemini 2.5 Pro)

Selected for its balance of advanced Japanese-language context understanding and generation speed, handling long prompts and image input in a single pass.

A 20,000-Character “Flame Director” Prompt

We verbalized the pile-on patterns unique to Japanese social media and consolidated them into one prompt, tuned to output an absurdity that entertains the user rather than hurts them.

Double-Layered Filtering

A generator AI and a moderator AI run in series and in parallel, screening every output. Personal attacks, comments on appearance, and anything beyond the danger line are mechanically removed.

Results & Impact

With zero paid media and purely organic spread, the campaign surpassed 1,000,000+ plays in its peak five days, 100M+ total impressions, and 10M+ views of related videos. Users shared their own unreasonable firestorms, and timelines overflowed with reactions.

Case Films

The full picture of the campaign is captured in the case films (approx. 2 min).

Voices from Players

All I posted was a photo of my cat, and I got slammed with “the houseplant in the background is toxic to cats — this is animal abuse.” The resolution of that internet-specific, self-righteous nitpicking gave me chills.

I always assumed online flaming had nothing to do with me, but watching my notifications flood with criticism felt so real my hands were literally shaking. So this is what “anyone can become a target” means…

Spread by Streamers and Creators

On day two, gaming streamers picked up the experience and the momentum flipped overnight. Reaction videos kept appearing unprompted, pushing total views of related videos past 10 million.

Beyond Promotion: A Crisis Simulator

The project began functioning beyond film promotion — as a “crisis simulator” reflecting a real social issue.

Students can safely experience how online firestorms work and why they are frightening. I want to use this as digital-literacy teaching material. — Educator

In an era where a small mistake can blow up overnight, could we use this as a reputation-risk training tool for employees? — Corporate risk manager

A single film promotion sparked a public conversation about flame culture itself.

Behind the Scenes

Pivoting Away from “Flame Video Generation”

The initial proposal — a generator of tabloid-style “scandal videos” — had already been approved. But CD Sakiko Koyama sent it back: “This is too weak as an experience to carry the film’s worldview.” That rejection sparked the concept of “Y,” a participatory simulation where the user becomes the target. Because resubmitting an idea this provocative would never pass on words alone, CEO Aoi Serikawa built a working AI mock in a matter of days, presenting feasibility and safety as a set — and the project restarted.

The Unglamorous Tuning of a Convincing Firestorm

The starting point for the “flame director” prompt was an online “flame quiz.” To teach the AI the mechanics of pile-ons, we fed it large volumes of mundane scenes — how someone cuts sausages, an ordinary family photo — training it to scan for “anything attackable” even where no fault exists. The CD and engineers met daily to polish the quality of each generated firestorm.

Not “What to Say” — “What Never to Say”

The hardest part of development was drawing the safety line. Outright character attacks like “unfit parent” or remarks about appearance only hurt people and carry no craft. The absurdity that works as entertainment had to be confined to deliberate exaggerations of fact and complaints about utterly irrelevant details — the kind that make users laugh at why anyone would care. A two-stage check by the generator AI and the moderator AI enforces this boundary mechanically.

Creative Control Through Prompting

Launch day was quiet. On day two a streamer picked it up and the mood flipped; by day five the experience had blown past one million plays. Reflecting on the result, CEO Aoi Serikawa put it this way:

With prompts alone, AI has already reached the point where creative output can genuinely be controlled.

The result was a promotion that let audiences live inside the film’s worldview before they ever entered the theater.