AI Video Generation Pipeline for Phiads
~$1.35 per AI-generated video
THE CHALLENGE
Phiads needed to produce UGC-style marketing videos at scale, but traditional video production was expensive, slow, and impossible to automate. Each video required a photographer, editor, and voiceover artist — costing hundreds of dollars and taking days per asset. The team needed a fully automated pipeline that could generate professional marketing videos from just a reference image and product description, at a fraction of the cost.
OUR APPROACH
We built a serverless AI video generation platform using a 5-step AWS Step Functions pipeline: Scene Analysis with Google Gemini 2.0 Flash to analyze reference images, Image Generation with Gemini 2.5 Flash to create UGC-style marketing photos, Script Generation to produce optimized 12-15 word video scripts, Video Generation with Kling AI 2.5 Turbo for 10-second clips, and Voice Sync using Gemini TTS with Sync.so lip-sync for the final product. The entire infrastructure was built with AWS CDK, featuring 11+ Lambda functions, DynamoDB, S3, and API Gateway with automatic 24-hour temp file cleanup.
THE RESULTS
Video production cost dropped to approximately $1.35 per video — a fraction of traditional production costs. The fully automated pipeline eliminates the need for photographers, editors, and voiceover artists. Videos are generated end-to-end without human intervention. The serverless architecture scales automatically with demand and costs nothing when idle. Webhook-based processing handles Kling and Sync.so callbacks reliably, ensuring consistent output quality across thousands of generated videos.