This is exactly what I built for a retail brand last month — a daily AI video pipeline running in production. Here is the architecture I delivered, and how it maps to what you need.
What I built for the client:
A scheduled n8n workflow that fires every morning, selects a product from a Shopify store using Claude API to write the script, opens HeyGen via browser automation to generate the avatar video, and then posts the output to Meta (Facebook + Instagram) at 10AM. The whole thing runs unattended daily.
How this maps to your requirements:
Script intelligence: Claude API takes competitor content and product context and generates the video script. I have already built this prompt chain for a retail use case and can adapt it to your brand voice and competitor angle.
Video generation: I used HeyGen for avatar-based video. Veo / Google Labs can slot in at the generation layer — the n8n workflow just changes which API call it makes. I can build it to support multiple backends.
8-second increment transcription: Handled via a JavaScript Code node in n8n — split the transcript by duration markers, align to video timestamps, output ready chunks.
Google Drive output: Google Drive API is straightforward in n8n — finished videos land in the right folder structure automatically.
Multi-account support: With 10+ social media accounts, I would build per-account scheduling logic in n8n so each account gets its own queue, rate limiting, and error handling without cross-contamination.
Why this is different from most replies you will get:
I do not need to figure out the architecture. I ran this in production for a real client. The pipeline has error handling, retry logic, and a daily canary check. I can show you the workflow structure on a call.
I am based in South Africa, fully remote, immediately available. DM me or reach me at narotsuhail@gmail.com.