Opus 5.5 把车内口播原片自动剪成 10 条成片
作者只给一段未剪辑的车内讲话视频,Opus 5.5 用多子代理分工完成 B-roll、剪辑、运镜、字幕,并自评打分循环返工,一次跑出 10+ 条可直接发布的短片。
I didn't edit this clip. Opus 5.5 did. I gave it 1 raw, uncut video of me talking in my car. It came back with 10+ coherent, ready-to-post clips. What it handled: • B-roll from my last livestream and my website • Cuts, zooms, and pacing • Closed captions synced to audio • A few other editing passes Here's the prompt structure I gave it: • Opus 5.5 acts as the orchestrator and breaks the edit into separate tasks • Each task is assigned to its own subagent (B-roll, cuts, captions, etc.) • The subagents complete their work and hand it back to Opus • Opus reviews and scores every output out of 10 • Anything below 9/10 goes back to that subagent with notes, and the loop repeats until every agent hits 9/10+ 1 run. No manual editing. Not perfect. The transition sound effects are overused. But the fact that it got this far on its own is genuinely impressive. This is what agentic workflows look like in practice. Would you post this as-is, or does it still need a human pass? 👇