AI Shorts Production Pipeline

Could a single operator run a short-form content pipeline without being physically present to publish? The goal was to find out if an AI-assisted stack — voice cloning, AI avatar, scripted workflow — could solve the time problem that breaks most solo content operations. Two content types ran in parallel: fast-hitting AI news shorts for reach, and AI tool tutorials for practical utility. Both through the same production system.

ChatGPT Projects held the system rules. ElevenLabs handled voice cloning. HeyGen produced the AI avatar. Every stage had a built-in checkpoint — hook, voice quality, lip-sync, thumbnail match. Anything that failed got rewritten or cut. Around thirty minutes of operator time per video, two hours with the editor. Total turnaround: two and a half hours.

Not fully automating it. Tutorials at the time showed n8n workflows producing finished videos with no human in the loop. I chose not to build that. The failure modes were real — mispronounced words, lip-sync drift, broken realism. Each needed a human catch before the video shipped. The edit style — quick cuts, kinetic text, motion every two to three seconds — was quality compensating for the AI's limitations.

The channel stopped at 28 videos. The pipeline and skill set carried into the next thing. The portfolio this project produced was part of what led to full-time work with Hance AI — video production for TechCrunch 2025 in San Francisco, Lenovo, Ferrari Formula 1, and Intel. Hance AI ceased operations in July 2026. The work produced during that time remains public.

Two videos broke out during the test window. Both publicly viewable.

The full case study documents the methodology, the system architecture, the numbers, the problems hit along the way, and what I did about them. Available by request.

Thanks. I'll send the case study within 24 hours.