05E-commerce & AI

AI Video Generator

E-commerce Content Automation at Scale

Canva APIYouTube APIPythonERP Integration

Overview

Project Overview

An e-commerce retailer with 500+ SKUs needed professional video content for every product to stay competitive on YouTube and social media. Manual video production was producing 20–30 videos per week at significant cost. They needed a 10x increase in output without a 10x increase in budget.

The Challenge

What We Were Solving

The client had a large product catalog fed by an ERP system with rich product data — descriptions, specs, images, and pricing. But translating that data into engaging video content required a designer, a video editor, and a social media manager for every product. The process was expensive, slow, and inconsistent. New product launches were delayed weeks waiting for video content.

The Solution

How We Built It

We built a Python automation pipeline that reads product data from the ERP API in batches, generates AI-enhanced product descriptions using GPT, assembles branded video compositions using Canva's API (templates + dynamic data injection), renders the final video, and uploads directly to YouTube with auto-generated titles, descriptions, and tags — all without human involvement. The system can process entire product categories overnight.

Technical Approach

Engineering Decisions

ERP Data Pipeline

A scheduled Python job polls the client's ERP via REST API, extracts new and updated products, enriches data fields using GPT-4, and feeds a processing queue — ensuring the content pipeline always reflects the live catalog.

Canva API Video Assembly

We built a set of Canva design templates with dynamic data slots. The pipeline injects product name, imagery, key specs, and pricing into each template via Canva's API, then triggers a render — producing consistently branded video frames at scale.

YouTube Publishing Automation

Rendered videos are uploaded via YouTube Data API with auto-generated SEO-optimized titles, descriptions, and keyword tags derived from the product data. Playlists are managed automatically by category.

Quality Gate

A lightweight validation step checks each rendered video for completeness (duration, resolution, data accuracy) before publishing. Failures are flagged for human review without blocking the rest of the batch.