Gosh... wish I could upvote twice. Feels like we just gave a low cost (for now) chainsaw to anybody who wish they had a pocket knife then say "There, you can cut anything with that!" and somehow they forgot they can just buy some OK stuff from Ikea or a nice artisan. The need to "build" anything without taking a minute to know, not even the state of the art, whatever already exist out there and "fix" it by "personalizing" it is nuts.

Let's not "vibe code" anything when reliable solutions already exist!

someone was asking for a GUI, so not going to be an ffmpeg expert. likely the LLM would recommend ffmpeg anyway. plus you would run YOLO (or maybe CLIP) locally; it has been running on Android phones since 2020 at least. a Jupyter notebook would also give a quick and dirty GUI to visualize and document the solution. plus “motion detection” is probably not the full story, and any video will probably have artifacting that means you’d have to tune the motion detection algorithm or end up with a bunch of garbage artifacts/false positives in the end. also, sounds like the user isn’t looking for something long-running like Frigate. if the user isn’t familiar with Python and wants to do something downstream like sort the outputs or whatever, an LLM would help with that.

sure, programmatically, it’s not a difficult problem, but like it or not it can be solved by someone without an advanced CS degree with an LLM precisely because the problem is easy. no easily ready solution exists, but that doesn’t mean it can’t be done. “just use ffmpeg” to someone like my dad who might have the know how to install Linux but isn’t a programmer isn’t exactly the simple advice it sounds like.