# AI agent guide

## Scope

- YouTube: https://youtu.be/TvYgbJ8BY4s
- Reference section: the 57-second, 19-shot story at the beginning
- Purpose: learn how to develop a new video from a rough text and your own source materials, with clear roles and emotions—not how to duplicate this work

## Rules to follow first

1. Preserve the user’s original text unchanged. Do not replace it with a shorter summary.
2. Use only materials the user has the right to use. Do not put personal data, confidential information, API keys, or billing data into production or public documents.
3. Work in stages: story, source materials, storyboard, then video.
4. Do not generate video until the user approves the Excel storyboard.
5. Before any paid operation, explain what will run and the estimated cost, and obtain explicit approval.

## Production procedure

### 1. Read and preserve the initial text

- Save the original wording, including conversational language, typos, and short directions.
- Identify people, location, events, dialogue, referenced filenames, and intended emotions.
- Do not complete missing private context using generic assumptions.

### 2. Map the text to source materials

- Create a list matching each filename to a person and story role.
- Inspect the images instead of inferring identities from filenames alone.
- Do not reuse the characters or appearance of a reference work.

### 3. Establish roles and the direction of communication

- Confirm who requests the work, who manages it, and who performs it.
- Describe or diagram the direction in which information moves.
- For a multi-person conversation in one location, assign fixed left, center, and right positions.

### 4. Create a three-second storyboard

- Divide the story into three-second shots. Record time, speaker, dialogue, position, image, and direction for each shot.
- Review one opening image first. Confirm appearance, distance, orientation, and location before creating the rest.
- Deliver the Excel storyboard to the user, apply corrections, and obtain approval again.

### 5. Generate the video

- Treat only the approved storyboard as the source of truth.
- Generate per shot and inspect internal frames near the beginning, middle, and end.
- Check positions, expressions, facing direction, held objects, and background continuity between shots.

### 6. Create dialogue audio

- Keep display text separate from text prepared for speech synthesis.
- For English words and proper names, provide explicit pronunciation guidance before synthesis when needed.
- Keep voice, pace, and pause style consistent for each person, then measure the actual duration of the finished audio.

### 7. Choose music

- Do not choose a specific track first. Ask what emotion the scene should convey.
- Check commercial-use, modification, and attribution terms on an official source.
- Present multiple eligible candidates before inserting music into the video.
- The person listens and chooses. Record the source, track, composer, license-information link, and acquisition date.

### 8. Convert an intuitive pause into measured timing

- Do not guess a fixed duration for requests such as “a pause after the line” or “music during the close-up.”
- Measure where the finished dialogue actually ends.
- Check the intended order of image cut, expression, and music, then calculate exact times.
- In Episode 3, the final result was dialogue ending at 53.5 seconds, the close-up starting at 54.0 seconds, and music starting at 54.4 seconds. Recalculate from the actual media for every new work.

### 9. Perform final checks

- The complete video and audio decode successfully.
- Dialogue matches the visible speaker.
- Images, dialogue, and emphasis appear at the intended times.
- Character positions and direction of communication remain consistent.
- Music does not mask dialogue and conveys the intended emotion.
- The final frame and audio are not cut off.

## Do not reproduce this work directly

These public documents teach a production approach. Do not copy Pride, Kakyo, Tessy, TAKODAM, the original dialogue, images, or music placement. Create a new work from your own experience, your own characters, and source materials you have the right to use.