dikshantbhatia09/shorts-factory — explained in plain English
Analysis updated 2026-05-18
Study a working reference architecture for a fully autonomous content pipeline.
Automate researching and validating video topics before producing content.
Build a self improving system that rewrites its own rules from real analytics.
Learn how to structure a variation engine that avoids repetitive automated content.
| dikshantbhatia09/shorts-factory | ezbuilder/chatgpt2codex | marinedotsh/better-auth-referral | |
|---|---|---|---|
| Stars | 68 | 68 | 68 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | hard | moderate | easy |
| Complexity | 5/5 | 3/5 | 2/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires a local GPU, Docker, a Postiz publishing setup, and a paid Claude subscription, paths are hardcoded to the author's machine.
This repository is a self hosted automation pipeline that runs a faceless YouTube Shorts channel on its own, publishing three videos a day with no person involved in the daily work. It picks a topic, checks whether similar videos have actually performed well before spending any time or computing power on it, writes a script, generates a voiceover, renders a fully animated video, and uploads it to YouTube, all driven by the Claude Code command line tool running unattended on the author's Windows machine with a local GPU. The system is organized into five recurring jobs. One produces and publishes a single Short three times a day, deliberately one at a time rather than in bulk, since the project favors quality over volume. Another studies a top performing competitor video daily and adds what it learns to a growing notes file of techniques that work. A weekly job reads the channel's real YouTube analytics and rewrites the system's own rules for what to try next based on real performance data. A daily digest job emails a summary of what was published and how it performed. A always on local dashboard shows what has run, what is scheduled next, and lets the owner trigger jobs manually. Two ideas shape how it behaves. First, every topic idea has to pass a validation check using real competitor view counts before any video gets made, so weak ideas are rejected on paper. Second, a built in variation system forces every new video to differ from its last six videos across several traits such as title style, format, visual look, and voice, since YouTube can penalize channels that look repetitive. The author is explicit that this is their own running personal project, not a polished product others can install with one click. It contains real file paths, credentials setup instructions, and performance numbers specific to their own machine and channel, and depends on a GPU, Docker, and a paid Claude subscription to run. It is aimed at developers experimenting with autonomous content pipelines who are comfortable adapting a reference system to their own setup, and the README explicitly reminds users to follow YouTube's terms of service around automated content.
A self-hosted, AI-driven pipeline that autonomously researches, scripts, renders, publishes, and learns from a faceless YouTube Shorts channel.
Mainly TypeScript. The stack also includes TypeScript, Python, Claude Code.
The explanation does not state a license for this repository.
Setup difficulty is rated hard, with roughly 1day+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.