# Technical information

## Finished story

| Item | Value |
|---|---|
| Duration | 42 seconds |
| Shots | 14 shots, 3 seconds each |
| Video | 1280×720, 24 fps |
| Audio | AAC, 48 kHz, stereo |
| Storyboard review | 9 frames per shot, 126 frames total |

## Models and tools by stage

| Stage | Model or tool | Purpose |
|---|---|---|
| Planning and correction | OpenAI Codex | Organized the story, dialogue, shot plan, and regeneration scope |
| Original redesign | Codex and image generation | Replaced the visual design with a purple-centered palette, compact round silver head, original torso, and small oval shield |
| Excel storyboard | Python, `openpyxl`, Pillow | Reviewed three-second images, explanations, and nine internal frames |
| Shots S001–S003 and S005–S010 | Google Veo API `veo-3.1-lite-generate-preview` | Generated four-second clips and normalized them to three seconds |
| Shots S004 and S011–S014 | Google Veo API `veo-3.1-fast-generate-preview` | Generated important action and ending shots |
| Dialogue | Google Gemini TTS `gemini-3.1-flash-tts-preview` | Produced controlled Japanese dialogue and Osaka-dialect pronunciation |
| Ambience | NumPy and FFmpeg | Generated non-melodic local ambience |
| Audio repair | FFmpeg | Normalized each segment to AAC, 48 kHz, stereo before concatenation |
| Final edit | Python and FFmpeg | Joined 14 clips into the 42-second story |
| Verification | FFmpeg, Pillow, `openpyxl`, speech recognition, and human review | Checked internal frames, speech, characters, machine scale, background, and ending |

## Video models in the final version

| Range | Model | Count |
|---|---|---:|
| S001–S003 | Veo 3.1 Lite | 3 |
| S004 | Veo 3.1 Fast | 1 |
| S005–S010 | Veo 3.1 Lite | 6 |
| S011–S014 | Veo 3.1 Fast | 4 |
| Total | 9 Lite and 5 Fast | 14 |

Model names reflect the production record at that time. A new production must check the current validated Video Studio defaults and current provider information.

## Lessons carried forward

- Define the relationship and story function instead of copying a reference work's appearance.
- Lock the head, torso, shield, colors, scale, and character positions in the storyboard.
- Keep final speech and ambience separate from generated video audio.
- Normalize four-second generated clips to the three-second storyboard unit.
- Keep display text separate from speech-oriented text.
- Normalize codec, sample rate, and channel count before audio concatenation.
- Keep rejected images and incorrect character versions in production history with reasons.