I want an AI editor for my project videos

Say I spend four hours stripping and sanding the floor in a room, and I record the whole thing. I want to turn that footage into a 30-second video, a two-minute video, and a ten-minute video. I want an AI editor that can look at what happened and help put those together.

That could be half an hour of footage or eight hours. A lot of it will be repetitive, but there will also be progress, problems, and things worth showing. I want it to figure out which parts matter without me having to watch the entire recording again.

Three versions of the same project

For the 30-second version, I want to show where I started, a few moments of the work, and how it came out. The two-minute version has room for the main steps. In ten minutes, I can keep more of the explanation, what went wrong, and how I dealt with it.

Those need different editing decisions. Just speeding up the same four hours to fit each length probably isn’t going to make something I want to watch. The editor needs to recognize when something changes and when I’m just doing the same thing for another twenty minutes.

I could help it while I’m recording

One thing I was thinking about is using a phrase when something interesting happens. I could say ‘mark that,’ and a timestamped transcript would give the editor somewhere to look. It would need to look before the phrase too, because I’ll probably say it after whatever happened.

Talking through what I’m doing could help in the same way. If I say I’m switching to a finer grit, or that something didn’t work, that gives it context for what it’s seeing. I don’t want to have to narrate every second, but a few comments could help it find the useful parts.

What I’d try first

The approach I want to try is to scan the footage, get a timestamped description of what’s happening, and look more closely at the interesting sections. From there, it could put together three edit plans and use the original footage to make the videos. I’d want to reuse that first analysis instead of starting over for every version.

I’ve been looking into local models and hosted options. I haven’t picked one or tested this on my footage yet. Cost and processing time both matter, but first I want to know whether it can choose footage that actually explains the work.

A short sample with a few different stages would be a good place to start. Can it spot the changes, keep something useful from each stage, and make a two-minute cut that makes sense? If it can do that, then I’d try it on a longer recording.

This is still an idea. I haven’t started building it, but it’s something I’d like to have for the projects I want to share here.