What this case covers
An editing outsider referenced the ChatCut architecture, used Claude and Codex to build a production line from talking-head transcription, A-roll cutting, and storyboarding to animation, and tested the results with real publishing data.
Questions it answers
- How is the entire AI editing production line composed?
- Why not do storyboarding before the script is finalized?
- Which materials are not suitable for AI generation?
- How does Feishu review reduce communication costs?
- What does high views but low completion rate indicate?
Our observations
- Workflow results must be tested with publishing data, not just by whether the finished video looks polished.
- After the opening gains exposure, a low completion rate means the middle section and video length still need optimization.
- AI handles the manual labor, while humans handle judgments that affect the authenticity of expression.
From the author
It is not an editing software; it is a working process of 'humans only make judgments, AI does the manual labor.'