I shoot my family videos on a DJI Pocket 3. We travel a fair bit with my daughter and my wife, and my approach is mostly point and shoot and see what it comes up with.
Since I'm on a family trip and all that it entails including leaving my camera recording because I forgot to turn it off as my daughter is crying or I need to help carry some luggage. I end up with clips that are unusable so at the end of every trip I've got a folder full of clips that aren't edited in any way. And that takes time.
Clip from the final video Claude helped me produce, some yummy snacks in our latest trip to Korea.
In the past I'd load everything into Premiere and go through building a timeline by hand. It's super time consuming and honestly it's the worst part of the whole thing. Half the time I didn't bother editing at all and just dumped everything into one timeline, because who has hours for that. A one week trip to Japan generates something like four hours of footage. I'm not sitting through four hours of my own footage on a weekend.
What I've been waiting for is the holy grail: dump my files into a folder and let AI edit it.
I haven't found a good solution in the wild for that yet, especially for long form. Short form is a different problem and there's plenty out there for it. But home videos are the opposite of short form. I want to preserve as much as possible, not cut it down to a highlight reel.
The closest thing I found was auto-editor, but it leans on audio detection, and that doesn't work for the stuff I wanted gone from home videos. There's not much dialogue in a clip of my daughter running around a park.
So I built my own and what I ended up with does two things. It exports a full quality sequential timeline, because home video clips are usually already in the right order and I want to watch them that way. And it strips out the junk: the bits where I left the camera running, where it was sitting in my backpack, the accidental recordings.
I borrowed auto-editor's 0.2 second padding default and its idea of exporting to an editor rather than only rendering a file. The detection is my own, and it's visual rather than audio: brightness, detail, and motion measured against each clip's own range.
I'm releasing it as a GitHub repo in case it helps save anyone else's time: https://github.com/EMChamp/winnow
Use it as is for your own videos, or fork it and extend it. I'm sure there are other ways to approach this.
I had it export to DaVinci Resolve because that's my editing software of choice, and because it's free, so anyone can open the timeline without paying for a subscription. I'm sure for otheres there would still be a lot of manual edits (and that's why I made an option to export to Resolve) but it solved a lot of my pain points and it gave me my weekend back.
Top comments (0)