只在画面上圈出要改的地方、说一句话，AI就把25分钟的播客剪得有模有样

Edit [your podcast recording filename] into a [target length]-minute [16:9 widescreen or 9:16 vertical] video for [your audience]. The source is about [source length] minutes. Focus on [main topic or purpose]. Keep each speaker's meaning, caveats, and natural delivery. Do not fabricate speech or use a cut that changes a claim. Keep the speakers visible for most of the episode. Preserve their names and roles: [names and roles]. First transcribe the full recording, or use the attached timed transcript. Check important names and numbers. Propose the sections to keep and cut, with source timestamps. Wait for my OK before changing the spoken edit. After approval: - Remove repeated points, false starts, and distracting dead time. Keep pauses and reactions that affect the meaning or mood. Do not shorten every breath. - Make clean audio edits that do not cut through words. Avoid unnecessary transitions between ordinary cuts. - Add titles at section changes and quotes or diagrams where they explain a point. Use [your brand colors] and [your font]. Keep text away from faces. - Use real archival footage only where it clarifies a specific point. Start with my supplied clips and sources. If search is available, propose usable real footage with its source and usage terms before adding it. - Do not substitute generated footage for requested archival footage. If footage is unavailable, keep the speaker visible and list the gap. - Keep speech clear. Add music only if I supplied it or approved the track. Save a cut list and an asset list in NOTES.md, including source timestamps, footage URLs, known usage terms, and where each asset appears. Work in sections if the episode is too long for one pass. Show me the preview. After my approval, run the project checks and export an MP4.

Cinko把试过的提示词整理成16条，在GitHub上免费公开，这是其中一条。按提示词包的说明，已与官方文档核对过，但没有原样用这段文字重新生成视频来测试。把方括号里的内容换成你自己录音的信息即可。把资料画面换成真实影像的后续提示词也在同一个文件里。