Why did I want to test AI video editing?
Video editing usually takes a lot of time, especially when you need to find the key moments in long footage, remove dead air, and adjust the pacing. The point of this test was to see whether AI video editing could reduce the amount of manual editing while keeping the content intact.
This video documents my hands-on test of using Codex to help edit a video.
At first, I only wanted to review my previous Hermes Agent tutorial and see where the edit could flow more smoothly. I ended up finding that Codex could do more than analyze the video itself: it could also help organize assets, identify edit points, work with subtitles, fix audio sync, and even add an outro.
Results from using Codex to recut a YouTube video
I also ran into some real-world problems during the test, including:
-
Videos that already have subtitles are not easy to recut directly
-
The first recording stopped unexpectedly and had to be joined with a replacement segment
-
ASR transcripts still needed manual correction
-
Subtitle position, the outro, and music all needed repeated adjustment
Overall, I think Codex is very interesting as part of a video editing workflow. It is not yet a complete replacement for editing software, but it can already take care of many tedious steps, especially editing tasks that are more engineering- or process-oriented.
If you are also experimenting with AI for video, subtitles, or automated editing, feel free to reach out and discuss it.