levelsio says one prompt to Claude Code edited his whole vlog

levelsio says his latest vlog was fully edited by Claude Code running Opus 5.5. He copied the clips off his DJI into a folder, opened Claude Code in that folder and asked it to cut them into a 10 minute vlog for X and YouTube. He says he made a few adjustments afterwards, but not a lot.

The interesting part is what the model built on its own to get there. By his account it transcribed the audio with mlx-whisper, Whisper large-v3-turbo running on his MacBook Pro GPU through Apple's MLX. It pulled frames every few seconds into contact sheets, grid images, so it could actually see what was in each clip. It used SpeechBrain ECAPA, a voice-matching model, to work out which voice was his and which belonged to other people. It generated thumbnails with OpenCV and YuNet's face detector. The edit itself ended up as a JSON file listing cuts, which clip, start time, end time, chosen by reading the transcripts and looking at the frame grids. He added on screen captions with Pillow, Python's image library.

He only found out any of this happened because Sam Parr asked him how he edited it.

The business read

This is a distribution story more than a tooling story. levelsio says that without AI editing he would not be vlogging at all, because the editing would be painful and slow for him and hiring someone would be more hassle. That is the thing worth noticing. The cost that kept a solo founder off a whole channel was not the camera or the idea, it was the three hours in a timeline afterwards. The vlog in question is the one where he worked through what is left when AI builds everything, so the format is already doing work for him.

Two caveats before anyone copies it. This is one founder's result on one video, not a benchmark, and he is the first to say he made adjustments. And the pipeline he describes is specific to his setup, Apple silicon, local models, Python libraries he did not pick himself. He did not ask for mlx-whisper or SpeechBrain. The agent reached for them.

What generalises is the shape of the ask. One sentence describing the finished artefact, a folder of raw inputs, and no instructions about method. That is a different way of using a coding agent than most founders are used to, and it only works when you are genuinely indifferent to how the job gets done.

@levelsio

@levelsio

X

All I asked for was "Claude can you edit these clips into a 10 minute vlog for X and YouTube

@levelsio

@levelsio

X

I think if I didn't have AI to edit it, I wouldn't be vlogging because editing would be such a painful and slow process for me