Make-Your-Own Legend Generator

Challenge

The film “The Beethoven Fabrication” (written by Bakarhythm, directed by Kazuaki Seki) is based on a true story: many of Beethoven’s legendary feats were fabricated by his secretary, Anton Schindler. Posters and trailers can tell you “great figures are manufactured” — but for the audience it stays someone else’s story. To deliver the film’s core, viewers had to become the protagonist of the premise themselves.

Approach

We built the “Make-Your-Own Legend Generator.” Enter your name and occupation, choose the image you want to leave to posterity, and upload a photo of your face. Seconds later, three outputs starring you are ready.

Technology & Ethics

Under the hood, the system runs on a two-tier pairing of a generator AI and a verifier AI, each given a different character so that creativity and reliability are handled as separate jobs.

Generator — Gemini 2.5 Pro

Chosen for its balance of creativity and Japanese-language context understanding. Configured so that “100 runs return 100 different answers,” it sits on the side that draws out wild, unexpected feats.

Verifier — Gemini 2.5 Flash

Chosen for cost-performance and configured to “return the same answer 100 times out of 100.” It checks for NG expressions, length limits, and brand-damage risk — mechanically stopping the generator from going off the rails.

A Copywriter-in-the-Loop Feedback Cycle

Through an admin console, a copywriter left line-by-line feedback on the generated copy — concrete notes like “this breaks as Japanese” or “this has nothing to do with the occupation.” Over 100 critiques were accumulated and folded back into the prompt, across more than 10 rounds of tuning. The result was not copy written by AI, but copy written by an AI steeped in a copywriter’s sensibility.

Cost & Operations

• Bulk generation runs on the Google Batch API, cutting API costs by 50%
• Speech synthesis runs on Cloud Run, auto-scaling with traffic spikes

Behind the Scenes

Turning the Briefing Conversation into a Proposal — On the Spot

The opening move: we generated the concept during the client briefing itself. The conversation was transcribed in real time, fed to Gemini, and a proposal was generated on the spot. The moment the meeting ended, we shared the screen: “Actually — based on what you just told us, we have a proposal.”

“You’re the first ones to ever pitch us on briefing day,” the client told us, visibly surprised. — Koji Okada (PR / Planning / PM)

By sharing a common scheme with “NOT ME THAT WENT VIRAL,” which was being proposed in parallel, costs came down and both projects were won together. The speed of the proposal changed how the work itself was won.

Possessing the AI with a Copywriter

The key was re-forging a copywriter’s know-how into the prompt. At first, the generator produced grammatically fine but unfunny output. The copywriter wrote down why this specific line isn’t funny more than 100 times in the admin console, and each batch was re-prompted. After 10+ rounds of tuning, the generator’s output shifted into the voice of an AI that had absorbed a human copywriter.