AI-Powered Automated Short Video Production Pipeline

Automated pipeline that turns products into videos

An automated pipeline that transforms a product into a ready-to-publish 9:16 vertical short video. AI writes scripts following a sales structure, converts them to natural voiceovers, generates multi-scene images, creates motion for each scene, then assembles everything into a complete video — no manual editing required.

The system allows cost-based quality control: lightweight previews for quick approval, standard quality for mass production, premium quality for flagship videos. Voice and characters remain consistent with each channel’s brand, avoiding the problem of every video having a different feel.

Sales-structure script generation

LLM writes content following the 4-part structure: Hook · Problem · Solution · Call to Action, optimized for short videos.

Text-to-speech conversion

Multiple voice providers (Google, ElevenLabs, Grok); add multiple voices per provider, select by gender and brand.

Multi-scene image generation & motion creation

Generate image sets for each scene, then convert to motion (image-to-video), assembling into 9:16 vertical videos.

Three quality levels, cost optimization

Preview 480p · standard 720p · premium 1080p; automatic quality downgrade when approaching budget.

Character & voice consistency

Lock “character identity” and voiceover to match channel brand, maintaining consistency across video series.

Preview before spending

Listen to voice samples and review lightweight drafts before rendering high-quality versions.

Per-AI-call cost tracking

Transparent cost ledger by stage (script, voice, images, video), attributed to individual videos.

Technical specifications

Specification	Value
Output format	9:16 vertical video (short video)
Script generation LLM	Qwen (primary) · GPT-4o (backup)
Text-to-speech (TTS)	Google Wavenet/Neural2 · ElevenLabs · Grok
Image & video generation	Wan 2.x — 480p / 720p / 1080p
Script structure	Hook · Problem · Solution · CTA
Backend	Node.js 20 · Fastify 5 · TypeScript
Queue processing	BullMQ · Redis
Storage	Cloudflare R2 (S3-compatible)
Cost control	Per-call cost ledger · 3 quality levels