| name | reels-playbook |
| description | Extract video (Reels/TikTok/Shorts/any MP4), analyze frame-by-frame, generate production playbook, then reconstruct in Remotion via video-director skill. Use when asked to deconstruct, clone, or recreate a video programmatically. |
| allowed-tools | Read, Write, Edit, Bash, Glob, Grep, AskUserQuestion, Task |
Reels Playbook → Remotion Reconstruction
Extract any short-form video, analyze frame-by-frame, generate a production playbook, then feed into video-director to reconstruct in Remotion.
Pipeline
Video source → Download → Frames (2fps) → Audio → Transcribe → Vision analysis → PLAYBOOK.md → video-director → Remotion code → .mp4
When to Use
- User has a video URL (Instagram, TikTok, YouTube Shorts, any)
- User has a local video file (.mp4, .mov, .webm)
- User wants to "clone", "recreate", "rebuild" a video in code
- User wants to understand how a video was made (decupagem)
- User provides a reference video for a new project
Input Types
1. URL (Instagram/TikTok/Shorts)
bash .claude/skills/reels-playbook/scripts/extract-reels.sh \
"https://www.instagram.com/reel/ABC123/" \
"playbooks/topic_name" \
2
2. Local Video File
Skip download, extract directly:
OUTPUT_DIR="playbooks/my_video"
mkdir -p "$OUTPUT_DIR/frames"
cp /path/to/video.mp4 "$OUTPUT_DIR/video.mp4"
ffmpeg -i "$OUTPUT_DIR/video.mp4" -vf "fps=2" "$OUTPUT_DIR/frames/frame_%03d.jpg" -y
ffmpeg -i "$OUTPUT_DIR/video.mp4" -vn -acodec pcm_s16le -ar 16000 -ac 1 "$OUTPUT_DIR/audio.wav" -y
Output Location
All outputs: playbooks/{slug}/
playbooks/{slug}/
├── video.mp4 # Original video
├── audio.wav # Extracted audio (16kHz mono)
├── transcript.txt # Timestamped transcript
├── metadata.json # Source metadata (if from URL)
├── frames/ # High-density frames (2fps)
│ ├── frame_001.jpg
│ ├── frame_002.jpg
│ └── ...
└── PLAYBOOK.md # Production playbook
Analysis Workflow
Step 1: Extract
Run script or manual extraction (see Input Types above).
Step 2: Read Context
- metadata.json → engagement, author, caption
- transcript.txt → what is said, timestamps
- ffprobe video.mp4 → duration, resolution, fps
Step 3: Analyze Frames (CRITICAL)
Read EVERY frame image. At 2fps, each frame = 0.5s.
For each scene/segment, identify:
Shot Composition:
- Wide / Medium / Close-up / POV
- Camera angle, static vs motion
- Framing, rule of thirds
Visual Elements:
- Background (color, gradient, image, video)
- Text overlays (content, font, position, animation, color)
- UI elements (cards, buttons, mockups, frames-within-frames)
- People/subjects, props
Motion & Transitions:
- Cut type (hard, jump, match, morph)
- Effects (zoom, pan, swipe, fade, wipe)
- Timing relative to audio/beat
- Pacing (cuts per minute)
Production:
- Color grading (warm/cold/natural/saturated)
- Split screen, picture-in-picture
- Overlay effects (grain, blur, glow)
Step 4: Generate PLAYBOOK.md
Structure:
# PLAYBOOK: [Topic/Creator]
> Decupagem de [duration]s — [source]
**Source:** [URL or file]
**Duration:** [X]s | **Resolution:** [WxH] | **FPS:** [N]
---
## METRICAS DE EDICAO
| Metrica | Valor |
|---------|-------|
| Duracao total | Xs |
| Total de cenas | X |
| Cortes por minuto | X |
| Duracao media de cena | X.Xs |
---
## ESTRUTURA GERAL
| Secao | Tempo | Duracao | Proposito |
|-------|-------|---------|-----------|
| Hook | 0:00-0:03 | 3s | Capturar atencao |
| ... | ... | ... | ... |
---
## DECUPAGEM FRAME-A-FRAME
### HOOK (0:00 - 0:03)
| Elemento | Detalhe |
|----------|---------|
| Shot | Close-up |
| Fundo | #0a0a0a gradiente |
| Texto | "Title" — Inter Bold 72px, centro |
| Animacao | Spring scale-in, 0.5s |
| Audio | Riser → Impact no reveal |
| Transicao → | Fade 0.5s |
### CENA N: [Nome] (0:XX - 0:XX)
...
---
## ELEMENTOS DE PRODUCAO
### Tipografia
| Fonte | Uso | Peso |
|-------|-----|------|
| Inter | Titulos | 800 |
| Inter | Corpo | 400 |
### Paleta
- Dominante: #XXXXXX
- Acentos: #XXXXXX
- Background: #XXXXXX
### Audio
| Elemento | Detalhes |
|----------|----------|
| Musica | [estilo/BPM] |
| SFX | [categorias: ambient, riser, impact, whoosh, tick, typing] |
---
## FORMULA REPLICAVEL
0:00-0:03 → HOOK: [pattern]
0:03-0:08 → SETUP: [pattern]
...
---
## REMOTION RECONSTRUCTION NOTES
Decisoes ja tomadas pela analise (pra alimentar video-director):
- **Tipo:** [title card / promo / explainer / social clip]
- **Duracao:** [X]s
- **Formato:** [WxH] @ [fps]fps
- **Fontes:** heading=[Font Weight], body=[Font Weight]
- **Fundo por cena:** [lista]
- **Animacoes:** [tipos identificados]
- **Audio:** [categorias necessarias]
- **Transicoes:** [tipos identificados]
Estas decisoes pulam o Discovery do video-director — ir direto pro Storyboard.
Step 5: Reconstruct with video-director
After PLAYBOOK.md aprovado:
- Invocar skill
video-director
- Pular Fase 1 (Discovery) — brief ja extraido do PLAYBOOK
- Ir direto pra Fase 2 (Storyboard) usando a decupagem como base
- Seguir fluxo normal: Scene Builder → Sound Design → Polish → Render
A secao "REMOTION RECONSTRUCTION NOTES" do playbook vira o brief do video-director.
Frame Timestamp Reference (2fps)
frame_001.jpg = 0.0s
frame_002.jpg = 0.5s
frame_003.jpg = 1.0s
frame_010.jpg = 4.5s
frame_060.jpg = 29.5s
Formula: timestamp = (frame_number - 1) / fps
FPS Guide
| Conteudo | FPS recomendado | Motivo |
|---|
| Talking head, tutorial | 1 | Pouca mudanca visual |
| Reels/TikTok padrao | 2 | Bom balanco |
| Fast edits, gaming, memes | 3 | Muitos cortes curtos |
| Transicoes complexas | 4+ | Precisa mais granularidade |
Quality Checklist
Antes de finalizar PLAYBOOK.md:
Dependencies
ffmpeg + ffprobe (frame/audio extraction)yt-dlp (video download from URLs)HIKER_API_KEY in .env (optional, for Instagram metadata)GROQ_API_KEY or OPENAI_API_KEY in .env (transcription)- Vision model access (frame analysis)
Related Skills
- video-director — receives PLAYBOOK output, runs 6-phase pipeline to Remotion code
- remotion-video-toolkit — Remotion coding patterns used by video-director
