/video-content — Video Assembly Pipeline

Three-tier video pipeline from static slides to platform-ready video. ffmpeg bookends every tier — slicing at the start, post-processing at the end.

For three-tier details, see rules/three-tiers.md. For Remotion slide archetypes, see references/remotion-archetypes.md. For ffmpeg recipes, see references/ffmpeg-recipes.md.

ffmpeg Bookend Architecture

START (always): ffmpeg slice
  Input:  tall artboard PNG (e.g., 2160×26880 @ 2x) OR individual slide PNGs
  Output: individual slide PNGs at target resolution (1080×1920)
  Method: crop=W:H:0:Y per slide, scale to target

MIDDLE (tier-dependent):
  v1 Quick:    ffmpeg crossfade stitch
  v1.5 Enhanced: ffmpeg Ken Burns + audio
  v2 Full:     Remotion animate with spring physics

END (always): ffmpeg post-process
  - Audio mixing (background music, SFX from Remotion)
  - Two-pass H.264 encode (CRF 18, movflags +faststart)
  - Thumbnail extraction (best frame or slide 1)
  - GIF preview (first 3 seconds, 480px wide)
  - Platform-specific encode (TikTok, Instagram, YouTube presets)

Workflow

Phase 1: Detect Input

Check for these artifacts (in order of preference):

1. Handoff YAML at marketing/handoffs/{name}-handoff.yaml — written by /paper-marketing
2. Content spec YAML at marketing/content-specs/{name}.yaml — written by /slideshow-script
3. Exported PNG — user provides path to tall artboard export
4. Slide directory — user provides path to individual slide PNGs

If handoff YAML exists, read it for:

• export_search_pattern — filename pattern for the exported PNG
• artboard.width, artboard.height, artboard.slide_count — for ffmpeg slice coordinates
• brand_snapshot — full palette + fonts for Remotion brand.ts generation
• content_spec — path to the content spec YAML for slide types and animation hints
• extracted_jsx — optional HTML/CSS per slide for pixel-perfect Remotion reference

Both handoff YAML and content spec YAML are needed for v2 Full tier. Handoff provides export/brand info, content spec provides slide content and animation hints.

If no artifacts exist, ask the user:

"What slides should I turn into a video?"

Options:
1. "I have a Paper export PNG" — provide path
2. "I have individual slide PNGs" — provide directory
3. "Run /paper-marketing first" — go back to design phase

Phase 2: ffmpeg Slice (START bookend)

If input is a tall artboard (not individual slides):

Where does SLIDE_COUNT come from?

• From artboard.slide_count in handoff YAML (preferred — exact value)
• From content spec YAML slides array length
• If neither exists, ask the user: "How many slides are in this image?"

# Detect slide count from image height
SLIDE_HEIGHT=$((IMAGE_HEIGHT / SLIDE_COUNT))

# Slice each slide
for i in $(seq 0 $((SLIDE_COUNT - 1))); do
  Y=$((i * SLIDE_HEIGHT))
  ffmpeg -i input.png \
    -vf "crop=${IMAGE_WIDTH}:${SLIDE_HEIGHT}:0:${Y},scale=${TARGET_W}:${TARGET_H}" \
    slides/slide_$((i + 1)).png
done

If input is already individual slides, skip to Phase 3.

Phase 3: Select Tier

Use AskUserQuestion:

"Which video tier?"

1. "v1 Quick" (~5 seconds) — ffmpeg crossfade, no animation. Good for draft review.
2. "v1.5 Enhanced" (~15 seconds) — ffmpeg Ken Burns + background audio. Good enough to post.
3. "v2 Full" (~90 seconds) — Remotion animated with spring physics, SFX, light leaks. Production quality.

Phase 4: Assemble (tier-dependent)

#### v1 Quick — ffmpeg Crossfade

# 3.5s per slide, 0.5s crossfade, H.264 CRF 18
ffmpeg -loop 1 -t 3.5 -i slide_1.png \
       -loop 1 -t 3.5 -i slide_2.png \
       ... \
       -filter_complex "xfade=transition=fade:duration=0.5:offset=3.0,..." \
       -c:v libx264 -crf 18 -pix_fmt yuv420p \
       -movflags +faststart output_v1.mp4

#### v1.5 Enhanced — Ken Burns + Audio

# Ken Burns: zoompan with subtle zoom + pan per slide
ffmpeg -loop 1 -t 4 -i slide_1.png \
  -vf "zoompan=z='min(zoom+0.0015,1.3)':x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)':d=120:s=1080x1920:fps=30" \
  ...
# Then stitch with xfade + add background audio

See references/ffmpeg-recipes.md for full Ken Burns and audio mixing commands.

#### v2 Full — Remotion Animate

1. Scaffold Remotion project (if not exists):

   mkdir -p marketing/video/{name}
   cd marketing/video/{name}
   bun init -y
   bun add remotion @remotion/cli @remotion/transitions @remotion/google-fonts @remotion/layout-utils

1. Read content spec YAML for slide content + archetypes

1. Generate slide components using archetypes from references/remotion-archetypes.md:

• Map type: stat → StatSlide component
• Map type: anchor_word → AnchorWordSlide component
• Map type: emotional_pivot → EmotionalPivotSlide component
• etc.

1. Auto-include in every v2 project:

• SFX on transitions: @remotion/sfx (whoosh, switch, page-turn mapped to archetype)
• Light leaks at emotional pivots: @remotion/light-leaks with hueShift for brand color
• calculateMetadata for dynamic duration based on slide count
• Zod parametrizable schema (colors, text editable in Studio)
• fitText() from @remotion/layout-utils for auto-sizing hero text

1. Opt-in features (ask user):

• ElevenLabs voiceover + auto-subtitles
• Custom SFX beyond defaults

1. Render:

   bunx remotion render src/index.ts TikTokSlideshow output_v2.mp4

Read the remotion-best-practices skill at ~/.claude/skills/remotion-best-practices/SKILL.md for animation rules, and load specific rule files as needed (transitions.md, timing.md, text-animations.md, etc.).

Phase 5: ffmpeg Post-Process (END bookend)

Always runs, regardless of tier:

# 1. Audio mix (if background music provided)
ffmpeg -i video.mp4 -i music.mp3 \
  -filter_complex "[1:a]volume=0.15[bg];[0:a][bg]amix=inputs=2" \
  -c:v copy mixed.mp4

# 2. Two-pass encode for optimal quality/size
ffmpeg -i mixed.mp4 -c:v libx264 -b:v 4M -pass 1 -f null /dev/null
ffmpeg -i mixed.mp4 -c:v libx264 -b:v 4M -pass 2 -movflags +faststart final.mp4

# 3. Thumbnail (slide 1 or best frame)
ffmpeg -i final.mp4 -vf "select=eq(n\,0)" -frames:v 1 thumbnail.png

# 4. GIF preview (first 3 seconds, 480px wide)
ffmpeg -i final.mp4 -t 3 -vf "fps=12,scale=480:-1" preview.gif

# 5. Platform-specific encode
# TikTok: H.264, 1080x1920, 30fps, CRF 18, AAC 128k
# Instagram: H.264, 1080x1350 or 1080x1920, 30fps, max 60s
# YouTube: H.264, 1080x1920, 30fps, CRF 16 (higher quality)

Phase 6: Output and Register

Output directory: marketing/video/{name}/

marketing/video/lumi-aida/
├── slides/           # Individual slide PNGs
├── output_v1.mp4     # Quick version (if made)
├── output_v1.5.mp4   # Enhanced version (if made)
├── output_v2.mp4     # Full version (if made)
├── thumbnail.png     # Best frame
├── preview.gif       # 3-second preview
└── src/              # Remotion project (v2 only)

Update brand/assets.md with new video asset entry.

Report:

Video ready:
  File: marketing/video/{name}/output_{tier}.mp4
  Duration: {seconds}s
  Size: {MB} MB
  Resolution: {width}x{height}
  Thumbnail: marketing/video/{name}/thumbnail.png
  Preview GIF: marketing/video/{name}/preview.gif

Standalone Usage

This skill works independently:

• /video-content alone — provide any slide PNGs, get video
• /paper-marketing → /video-content — design + assemble (no scripting)
• /slideshow-script → /video-content — skip Paper, use existing images
• Any PNG source → /video-content — works with Canva exports, screenshots, anything

Prerequisites

• ffmpeg installed (8.0+)
• bun installed (for Remotion v2)
• For v2: Remotion packages installed automatically during scaffold

Anti-Patterns

• Wrong pixel format — Always use -pix_fmt yuv420p for H.264. Without it, some players show a green screen or won't play at all.
• Missing faststart — Always include -movflags +faststart for web/social video. Without it, the video won't play until fully downloaded.
• Scaling after stitching — Always scale individual slides BEFORE stitching. Scaling the final video degrades quality.
• Skipping the slice step — Even if the input "looks like" individual slides, verify dimensions. A tall artboard that isn't sliced will produce a single stretched frame.
• CRF too high — CRF 18 is the sweet spot. CRF 23+ looks muddy on mobile. CRF below 15 balloons file size for marginal quality gain.
• Assuming Remotion packages exist — Before importing @remotion/sfx or @remotion/light-leaks, check if they're real packages. If unavailable, implement SFX with standard <Audio> component and light leaks with CSS gradients + opacity animation.
• No audio track — Social platforms may reject videos without an audio track. For v1, add a silent audio track: -f lavfi -i anullsrc=r=44100:cl=stereo -shortest.

Edge Cases

• ffmpeg not installed — Check with which ffmpeg. If missing, tell the user: "Install ffmpeg: brew install ffmpeg (macOS) or apt install ffmpeg (Linux)." Do not proceed without it.
• Image height not divisible by slide count — Round down the slide height. Warn the user that the bottom few pixels may be cropped. This is usually imperceptible.
• Remotion render fails — Common causes: missing fonts (use @remotion/google-fonts), missing assets in public/, or TypeScript errors. Check the error output and fix before retrying.
• Disk space — Video rendering needs space. A 7-slide v2 project can use 500MB+. Check available space before rendering.
• Background music sourcing — For v1.5, the user must provide a music file. Suggest royalty-free sources: YouTube Audio Library, Pixabay Music, or Uppbeat. Never use copyrighted music.
• Corrupted PNG input — If ffmpeg fails on a specific slide, verify the PNG opens in an image viewer. Re-export from Paper if corrupted.

Principles

• ffmpeg bookends everything — slice at start, post-process at end, regardless of tier
• Tiers are progressive — v1 is instant, v1.5 is better, v2 is production. User chooses speed vs quality.
• Content spec drives everything — slide types, animation hints, and voice constraints come from YAML
• Remotion best practices — always load the remotion-best-practices skill for v2 work
• No hardcoded content — all text, colors, and timing come from content spec + brand files