YouIAautoA video factory that never switches off.
Six YouTube channels scripted, voiced, edited and published by one server on AWS.
Overview
A video factory that never switches off.
Six YouTube channels scripted, voiced, edited and published by one server on AWS.
Product definition and AI-assisted development, connecting requirements, implementation and review.
Running on AWS since October 2026: 6 channels publishing and 1 paused on purpose, 780 tests passing on 1 Nov 2026 and release v028 installed with no restart and no test uploads. The target of 5 Shorts + 1 long video per channel is still being measured.
The problem
Publishing daily on several channels takes topics, scripts, media, voice, editing and uploads, and a small server cannot do it all at once. A failure halfway must never turn into a duplicate video or a wasted quota.
The solution
One worker per channel behind shared locks: one heavy production at a time and coordinated uploads. An idempotent ledger records every upload attempt. Media search federates eleven providers and drops NC, ND or unknown licences; CC BY credits go into the description. Each channel has its own voice, pace and pronunciation dictionary.
Architecture
Topics — Official YouTube and Google Trends samples, filtered by niche.
An explanatory diagram of the running workflow. This is not live system monitoring.
Features
Automatic script, voice-over, edit and upload
Implemented¹Media from 11 providers with checked licences
Implemented¹Idempotent upload ledger
Implemented¹Daily target per channel being measured
In development¹ Identified in code or implementation documentation; not equivalent to a new production test.
Technologies
Technical challenges
Disk. The server has 30 GB and each production uses about 1 GB while it works. I added a reserve that delays production instead of crashing, staged installs that abort safely and a clean-up that only deletes videos already public on YouTube: 2.83 GB freed without touching scheduled items, the database or the queues.
Real screenshots
Real output produced by the system.
Lessons learned
A running container is not a goal that is met. I only call it stable after measured full cycles, and every install leaves a receipt saying what changed, what was tested and how to roll back.
Links
The Short shown uses images from Wikimedia Commons, Openverse, iNaturalist, StockSnap and rawpixel under CC0, public domain, CC BY 2.0/2.5/4.0 and CC BY-SA 2.0/3.0/4.0. Featured photos: “Where Baby Flamingos Come From” (Eric Kilby, CC BY-SA 2.0), “Pink flamingos – Cervia salt pans” (Wikimedia Commons, CC BY-SA 4.0) and a pigeon photo from StockSnap (CC0).
Public repository not provided.
Next conversation
Let’s build something.An interesting problem?
Let’s build.
Web systems, automation and tools that turn an idea into working software.