# Totally Spies S7 benchmark packs

## Purpose

This note defines three reusable benchmark packs built from the downloaded
official Season 7 corpus.

The goal is to make the earlier animation-style conclusion operational:

- validate the production style repeatedly against a broader sample
- separate continuity-style references from short-form benchmark references
- prepare practical packs for future training, evaluation, and prompt testing

## Pack root

All pack assets live under:

- `materials/benchmark/youtube-s7-validation/packs/`

Shared summary:

- `materials/benchmark/youtube-s7-validation/packs-summary.json`

## 1. Dialogue / hold pack

Path:

- `materials/benchmark/youtube-s7-validation/packs/dialogue-hold/`

Files:

- `manifest.json`
- `manifest.csv`
- `snippet-manifest.json`
- `snippet-manifest.csv`
- `still-sheet.png`
- `motion-sheet.png`

Count:

- **13** references
- **13** references with snippets and sequential motion frames

Selection logic:

- character-intro promos
- briefing / planning clips
- low-action explanatory beats

Selected references:

- `Meet Alex!`
- `Meet Clover!`
- `Meet Cyberchac!`
- `Meet Glitterstar!`
- `Meet Jerry!`
- `Meet Mandy!`
- `Meet Sam!`
- `Meet Toby!`
- `Meet Zerlina!`
- `The Future of Spy Missions?`
- `Training Against Simulated Agents!`
- `A New Mission in Singapore!`
- `The Spies Meet A New Team Member!`

Use this pack for:

- validating held-frame behavior
- testing dialogue and intro beats
- checking whether a model can preserve stable faces and bodies across low-
  motion sequences

## 2. Action / gadget pack

Path:

- `materials/benchmark/youtube-s7-validation/packs/action-gadget/`

Files:

- `manifest.json`
- `manifest.csv`
- `snippet-manifest.json`
- `snippet-manifest.csv`
- `still-sheet.png`
- `motion-sheet.png`

Count:

- **12** references
- **12** references with snippets and sequential motion frames

Selection logic:

- gadgets
- fights
- threat-response beats
- action-leaning clips and compilations

Selected references:

- `Brand New Gadgets!`
- `Best Gadgets!`
- `The Spies' Best Fights`
- `Stopping the WOOHP Express!`
- `Spacewalk In Style`
- `Wedding Makeup Mayhem!`
- `Fighting A Giant Durian Fruit!`
- `Fighting Digital Bugs!`
- `Bug Mayhem in the Camp!`
- `The Spies Save Singapore!`
- `A Frightful Night For The Spies!`
- `Totally Pest Control!`

Use this pack for:

- validating whether action remains restrained rather than becoming overly
  fluid or noisy
- testing gadget readability and attachment continuity
- checking whether the model preserves crisp 2D cutout-style motion under
  pressure

## 3. Character / costume consistency pack

Path:

- `materials/benchmark/youtube-s7-validation/packs/character-costume-consistency/`

Files:

- `manifest.json`
- `manifest.csv`
- `still-sheet.png`

Count:

- **16** references

Selection logic:

- main-character intros
- supporting cast intros
- hero best-of compilations
- trailer / intro / music assets

Selected references:

- `Meet Alex!`
- `Meet Clover!`
- `Meet Sam!`
- `Meet Jerry!`
- `Meet Zerlina!`
- `Meet Mandy!`
- `Meet Toby!`
- `Meet Cyberchac!`
- `Meet Glitterstar!`
- `Best Of Alex!`
- `Best Of Clover!`
- `Best Of Sam!`
- `Totally Spies! Season 7 Trailer`
- `Totally Spies! | Season 7 Intro Song!`
- `Totally Spies! Music Video | BRAND NEW Season 7`
- `Totally Spies! KARAOKE Music Video | BRAND NEW Season 7`

Use this pack for:

- face and silhouette consistency checks
- wardrobe / costume continuity checks
- validating whether generated outputs preserve the franchise-coded character
  identities cleanly across contexts

## Why these packs matter

Together, the three packs make the earlier conclusion testable in a practical
way:

- **Dialogue / hold pack** tests the limited-animation side directly
- **Action / gadget pack** tests whether even action remains controlled and
  cutout-like
- **Character / costume consistency pack** tests stable identity, costume, and
  silhouette preservation

## Recommended use order

1. `character-costume-consistency`
2. `dialogue-hold`
3. `action-gadget`

That order follows the likely approval risk:

- first lock the people
- then lock low-motion stability
- then test action / gadget pressure
