⭐ Storyboard MCP · Ultimate showcase
THE ROAD a history of human transportation
Ninety seconds. Eighteen era beats. Five-thousand-five-hundred-fifty years of human movement. One evolving road. Built end-to-end via the storyboard MCP on the Livepeer GPU network — Netflix-quality cinema cut and vertical TikTok cut, generated in parallel, mastered together, stitched and scored as a single artifact.
Cinema cut · 1920×108090.0s · 18 beats × 5s · stitched with unified score
Vertical · 1080×1920TikTok / Reels / Shorts
The through-line is the story. A road that ages from a dusty Mesopotamian path → Roman basalt slabs → medieval mud → cobblestone → asphalt → multi-lane freeway → smart-road with embedded LEDs → and finally lifts away to sky. The road is the protagonist. The vehicles are guests. The human eye holds continuity across cuts by anchoring to ONE persistent element — that's why this reads as time travel and not a slideshow.
The 18 beats — one keyframe per era, 5 seconds each in motion
Each beat was rendered as a dual-aspect keyframe pair via gpt-image (cinema 2.35:1 at 1536×1024, vertical 9:16 at 1024×1536 — native composition both ways, not center-cropped), then animated by kling-v3-i2v for 5-second cinematic motion. All 36 keyframes succeeded on the first try with zero moderation issues.
Pipeline trace — what the storyboard MCP actually did
Wave 1 — keyframes
36 × create_media action=generate, model=gpt-image, aspect_ratio=2.35:1 OR 9:16 → 36/36 done in ~111-148s each, all on first try
output: 18 cinema 1536×1024 PNGs + 18 vertical 1024×1536 PNGs, all fal.media-hosted
Wave 2 — i2v animations + score
36 × create_media action=animate, model=kling-v3-i2v, duration=5, source_url=<wave1 url> → 36/36 done in ~54-75s each, ZERO moderation flags
1 × create_media action=music, model=minimax-music, duration=15 (returned 115s, score builds across full reel) → done
Wave 3 — local stitch
ffmpeg normalize: scale+crop 1108×828 native → 1920×1080 (cinema), 1080×1920 (vertical)
ffmpeg concat: 18 clips → single timeline per format (concat demuxer, no re-encode)
ffmpeg mix: unified score @ 0.95 vol + clip audio ducked to 0.22, afade in/out
output: the-road-cinema.mp4 (72MB, 90.0s) + the-road-vertical.mp4 (74MB, 90.0s)
Quality controls applied
✓ Storyline approved before any MCP fires (18-beat treatment surfaced for review first)
✓ Same camera-rule across all 18 keyframes ("eye-level wide, 35mm equiv, subject from middle-distance")
✓ Through-line continuity ("the SAME road, now paved/cobbled/asphalt/smart-LED") in every prompt
✓ One-day lighting arc (frosty dawn → noon → afternoon → dusk → twilight → sunrise) — the earth turns once across 90s
✓ "Sign shapes only, no readable text" guard against AI-text rendering failures
✓ Dual-format native composition (gpt-image aspect_ratio param, NOT post-crop)
✓ i2v completion-emphasis ("Complete continuous forward motion through the entire clip")
✓ i2v moderation fallback chain armed (kling → seedance → ltx) — never triggered, all 36 passed
Cost — total MCP run end-to-end
Wave 1 keyframes
~$7.20
36 × gpt-image @ ~$0.20
Wave 2 animations
~$20.50
36 × kling-v3-i2v @ ~$0.57
Music score
~$0.40
1 × minimax-music
Total
~$28.10
on-target for the $25–35 flagship playbook
Reproduce this — the playbook is open
This reel was built from the history-journey playbook — the same template can produce a history-of-fashion (tailor's mirror as through-line), history-of-food (kitchen counter), history-of-music (piano keyboard), history-of-architecture (city horizon), or any other arc where ONE evolving visual element ties N era beats together. Paste the playbook into Claude with your YAML config filled in — same agent, same 4-wave pipeline, same quality controls, different subject.