The Pipeline: Five Stages
The goal is not full automation but freeing the human from repetitive work: topics accumulate automatically, scripts are AI-drafted, visuals are template-generated, distribution is scheduled across platforms, and analytics flow back for review. The human does exactly two things — final topic approval and key on-camera or voiceover judgment.
Tool Stack per Stage
| Stage | Tools | Monthly Cost |
|---|---|---|
| Topic sourcing | n8n scraping hot lists and RSS | 50 CNY |
| Script drafting | Kimi / ChatGPT | 0-99 CNY |
| Visuals | Dreamina + Kling + CapCut templates | 100 CNY |
| Voiceover | Moyin / ElevenLabs | 60 CNY |
| Distribution and review | Platform schedulers + Feishu dashboards | 0 CNY |
Total cost is about 300 CNY per month, with output comparable to a 4-person short-video team.
Three Quality Gates
- Topic gate: of 50 AI-suggested topics, the human keeps 10 — an 80% rejection rate;
- Fact gate: every script with data or citations gets manually verified — AI drafts carry roughly 5% factual errors;
- Platform gate: each video passes a sensitive-word and compliance check before publishing.
One Month of Production Data
In 30 days the pipeline produced 912 videos — 30 per day — with 24 million total views. Three percent of the videos drove 70% of the views. The creator spent about 2 hours daily, mostly on topic review and hit analysis. The lesson is clear: the pipeline solves volume; hits still come from human taste.
Advice for copycats: run the full process manually 20 times before automating anything. Automating a broken manual process only produces garbage faster.