Video Ad Pipeline
AI-Generated Video Advertising
An end-to-end AI pipeline that transforms creative briefs into production-ready video advertisements. The pipeline ingests a text brief (product, audience, tone, duration), generates a script with scene breakdowns using Claude, produces video segments via Runway ML, synthesizes voiceover with ElevenLabs, composites everything with motion graphics and brand overlays using FFmpeg, and delivers review-ready assets to a Next.js dashboard. Campaign managers can review, request revisions, approve, and export in multiple formats. A single brief can generate multiple variants (different lengths, tones, CTAs) in one pipeline run.
What Was Built
Brief Ingestion
structured intake form for product details, target audience, tone, duration, brand guidelines, and reference material. Claude generates a creative brief expansion with script and storyboard.
Script & Storyboard Generation
Claude produces scene-by-scene breakdown with visual direction, dialogue/voiceover text, on-screen text, and timing. Supports revision cycles before production.
Video Generation
Runway ML API for AI video generation per scene. Style transfer from reference material ensures brand consistency. Parallel generation of multiple scenes.
Voiceover Synthesis
ElevenLabs Professional voices with multilingual support. Automatic timing alignment with video segments. Multiple voice options per brief.
Composition Engine
FFmpeg pipeline for final assembly: video segments + voiceover + background music + motion graphics + brand overlays + lower thirds + end cards. Renders multiple formats (16:9, 9:16, 1:1).
Review Dashboard
Next.js app for campaign managers to review generated ads, provide scene-level feedback, request revisions, approve finals, and export in multiple formats and resolutions.
How It's Built
n8n orchestration for multi-step pipeline with error handling and retries
Parallel scene generation for faster throughput
S3 asset storage with CloudFront CDN for review delivery
FFmpeg with custom filter graphs for professional composition
Webhook-based status updates to review dashboard
Queue-based architecture — briefs processed asynchronously with progress tracking
Multi-format rendering: landscape (16:9), portrait (9:16), square (1:1)
Version control for revisions — full history of generated variants
Technical Stack
Results & Impact
Brief-to-ad pipeline runs in under 15 minutes for a 30-second spot
Multi-variant generation produces 3-5 ad variations per brief automatically
Multi-format rendering serves all major platforms (YouTube, Instagram, TikTok)
Review dashboard reduces approval cycle from days to hours