# TAKO and AI agent production conversation

These are public-safe excerpts from the real production decisions behind the 57-second opening story. The initial misunderstandings and corrections are retained rather than rewritten as a perfect success story.

## 1. The first draft had two people speaking directly

TAKO initially gave the AI a rough text in which Pride spoke directly to Tessy. The AI read the text and source materials and organized it into 19 three-second shots, totaling 57 seconds.

After seeing the first attempt, TAKO realized that the real relationship was different.

> Previously, Pride asked Tessy directly, but that was wrong.

> End-client information-systems representative: Pride
> PMO: Kakyoin, a new character
> Engineer: Tessy

The AI revised the relationship to Pride → Kakyo → Tessy.

## 2. Keep all three people in place and send the request right to left

The first visuals also made the positions and facing directions feel unnatural. TAKO clarified the name and staging.

> Can you change Kakyoin to Kakyo?

> It would be funnier if all three stayed lined up while the conversation continued.
> That would make the right-to-left feeling clearer, wouldn’t it?
> Do you see what I mean?

The AI fixed the command-room positions as Tessy on the left, Kakyo in the center, and Pride on the right. The request moves from right to center, then center to left.

This staging made the PMO’s simple forwarding of the request visible, not merely verbal.

## 3. The ending music was not decided at the beginning

There was initially no song. TAKO first asked for a hard-rock “OH NO” at Tessy’s shocked ending. After watching the result, he changed direction.

> About the “oh no” at the end—

> Could you find several tracks like the music from *Tales of the Unusual*—something that feels as if you have wandered into a strange space—and propose them before putting anything into the video?

The AI searched the royalty-free music site DOVA-SYNDROME, checked suitable candidates, and proposed four tracks. Its first recommendation was “Kimyouna Annainin.” TAKO chose it after listening.

> Let’s use the first one.

The selected track was “奇妙な案内人” (“Kimyouna Annainin”) by MATSU, distributed through DOVA-SYNDROME. The person described the desired emotion, the AI found candidates with checkable usage conditions, and the person made the final choice.

## 4. Deriving the final pause and music timing

TAKO did not specify an exact start time. He described the effect he wanted after watching the video.

> At the end, after Kakyo finishes speaking,
> there is a pause,
> then a close-up of Tessy,
> and the music starts during the close-up.
> Would that be better?

The AI inspected the finished dialogue audio and images and derived these times:

- 53.5 seconds: Kakyo’s line ends
- 54.0 seconds: cut to Tessy’s close-up
- 54.4 seconds: “Kimyouna Annainin” begins

The actual gap between the voice ending and the music beginning is about 0.85 seconds. TAKO supplied the emotional and visual intention; the AI converted it into exact timing from the finished media.

Tessy was also changed so that he looks confused from the first frame of the close-up, rather than becoming surprised partway through it, while the camera slowly moves closer.

## 5. What this process shows

- A rough first text can begin a storyboard when it includes people, source materials, events, and the intended feeling.
- The AI does not already know the private real-world context. A person must watch the attempt and correct mistaken roles and relationships.
- Fixed character positions and the direction of information flow can matter as much as dialogue.
- Music can be chosen by first describing the emotion, then comparing candidates and licensing conditions.
- An intuitive request such as a “natural pause” can be measured against the finished dialogue and video and converted into exact timing.

There were other corrections too, including characters appearing back-to-back, music entering too early, and Tessy’s expression. Some of these appear later in the landscape-format video’s failure compilation.