Case study 27 / 37
Animation Ad
An agent skill that reads a product's source, writes a storyboard and renders a real animated ad locally: motion graphics, characters or low-poly 3D, with voice, music and captions.
- Status
- Working build
- Domain
- AI · Tools
- Source of claims
- Private repository README (reviewed). Source is not public.
Run /animate-ad from a coding agent: it infers product details from the project, asks only for missing decisions, presents a storyboard, then renders on the CPU with FFmpeg. No API keys, GPU or Blender are required. Output includes MP4, captions in SRT/VTT and a technical QC report.
01/The problem
Product videos are expensive and usually disconnected from what the product actually does. An agent that can read the code can base the script on real features, and should never pretend a model did work it didn't.
02/The system
Run /animate-ad from a coding agent: it infers product details from the project, asks only for missing decisions, presents a storyboard, then renders on the CPU with FFmpeg. No API keys, GPU or Blender are required. Output includes MP4, captions in SRT/VTT and a technical QC report.
03/Scope
- 01Evidence-backed concept, script, storyboard and an editable scene plan authored by the agent.
- 02Original CPU-rendered motion graphics, 2D characters with stable identities, and low-poly 3D geometry.
- 03Local multilingual speech with audio-reactive mouths, synthesised music and transitions.
- 04MP4 with optional WebM/GIF, burned captions, SRT and VTT, version-preserving revisions and per-scene renders.
- 05Schemas for 13 languages including Nepali, Hindi and Arabic; formats from 9:16 to 16:9, 6 to 120 seconds.
04/Engineering
No silent substitution
A requested style the local renderer can't produce is never quietly swapped, and the CLI never claims a model authored a plan it didn't.
QC that knows its limits
Technical checks cover decoding, duration, dimensions, missing audio, silence and black frames; the report states what it cannot certify: effectiveness, cultural quality, lip-sync.
05/Interface
Interface screenshots of this commercial product are not public. The visual above is an abstract representation of its modules — not the product itself.
06/Tech stack
- Node.js 22
- FFmpeg
- Agent Skills
- eSpeak
- Procedural 2D / low-poly 3D
07/Result
Verified outcomes
- Six public regression briefs (SaaS, jewellery, restaurant, travel, 3D character, Nepali) with test, check and benchmark scripts.
Known limitations
- Local speech is basic synthesis; realistic characters and cinematic generative video need external capabilities.
08/Links
Private commercial codebase — no public links.
Related work
Versioned system-design workspace
ArchitectFlow →
Design a system's architecture, data model, API and UI as one versioned document, check it with deterministic rules, and generate an implementation prompt from the exact saved revision.
Open source · Experimental
Market tracker
Puffer →
A full-stack stock tracker with watchlists, market search and AI-written news digests delivered by background jobs.
Open source · In development
AI-text classifier
AI Slop Detector →
An end-to-end AI-writing detector: scraped data, synthetic generation, a LoRA-tuned BERT classifier, and an honest benchmark.
Open source · Research
Next case study
Puffer →