Sample library

Three fictional products, six markets, three formats. Pick a combination to see the script, end card, storyboard and predicted scores each market would get.

Product

Markets

Format

Botanica No.7 Night Serum

Skincare · 30 ml · 3 markets · 9:16

成分を、一滴ずつ設計しました。

公式ストアで見る · ¥9,800

● 00:00:00:0000:00:12:00 · 30 FPS

V1Runway Gen-4.5 B-Roll

S1Claude Script & Subtitles

A1Regional Voice Synthesis

E1Akamai Edge HLS Stream

Illustrative keyframe for a fictional product. Drag the playhead to scrub.

Japan · ja-JP

成分を、一滴ずつ設計しました。

7種の植物エキスを低温抽出。皮膚科医監修、無香料。

公式ストアで見る · ¥9,800

Back-translation: We engineered the formula, drop by drop.

Precision and provenance over urgency; softer CTA toward the official store.

Predicted before spend (model estimates, not results)
3-second hold score	92.4 / 100
Predicted thumbstop rate	31.8%
Predicted ROAS	3.1×
Served from	Tokyo-NRT

All selected markets

Market	Hook	Voice-over	End card
Japanja-JP	成分を、一滴ずつ設計しました。	7種の植物エキスを低温抽出。皮膚科医監修、無香料。	公式ストアで見る · ¥9,800
Germanyde-DE	Sieben Wirkstoffe. Null Kompromisse.	Kaltgepresste Pflanzenextrakte, dermatologisch getestet, ohne Duftstoffe.	Jetzt entdecken · 59 €
Brazilpt-BR	Seu skincare noturno em uma só gota.	Sete botânicos prensados a frio. Pele com brilho em sete noites.	Compre em até 6x sem juros · R$ 349

Storyboard

00:00:00:00
Macro: dropper meniscus, rim light
Slow push-in, 24mm, f/2.0
00:00:02:12
Hero: bottle on stone plinth
Orbit 18° left, rack focus
00:00:06:00
Texture: serum bead on stone
Locked-off, 120fps ramp
00:00:11:18
End card: localized CTA + price
Fade-up, 12f

What happens in a render

The four stages a design-partner render goes through, with an example of the log each one writes.

1
Claude Sonnet 4.5
Extracting SKU value props & transcreating regional hooks
- GET https://shop.example.com/products/botanica-no7-night-serum → 200 · parsed 14 product attributes
- brand_voice.extract → tone=precise, register=premium
- transcreate.hooks → 3 locales · glossary terms locked: 6
- camera_prompts.generate → 4 shots · motion vectors attached
2
Runway Gen-4.5 API
Synthesizing 60fps camera motion & regional scene lighting
- image_to_video.submit → keyframe serum-16x9.jpg · 4 shots
- scene.relight → 3 regional grades (LUT: cinema-warm-v3)
- act_one.presenter → lip-sync tracks queued · 3 voices
- render.complete → 4K / 60fps · 12.0s per variant
3
GCP Vertex AI
Computing frame-by-frame saliency & 3s hook retention score
- saliency.map → 720 frames sampled · product in focus 81% of hook
- retention.predict → 3s hook score 92.4/100
- bigquery.write → telemetry.campaign_variants (3 rows)
4
Akamai / Linode GPU Cluster
Running FFmpeg NVENC subtitle burn-in & HLS edge distribution
- ffmpeg -hwaccel cuda -c:v h264_nvenc · subtitles burned · 3 variants
- hls.package → 6s segments · 1080p/720p/480p ladder
- edge.publish → 4,100+ PoPs · cache warm