This is part two of my live build of a video recorder app with Claude Code. In part one I planned the app and got a first prototype that shows a screen share, a webcam and a changeable background. In this stream I make it a tool I would actually record with.

Where we started

The app already recorded a screen and a talking head on one canvas, and it could blur the background behind me. But the screen area was static. If it pointed at the wrong part of the screen, I could not fix it before recording, and that defeats the point.

The reference: my short form recorder

I had built a second recorder the day before for vertical short form video, and that one was close to what I wanted. It puts the screen recording in the top half and the face cam in the bottom half, with a divider line between them. You can zoom and move each part while recording, change the split (for example 50% screen and 50% cam), and swap the background or upload your own. It also shows an audio waveform, so you can see that your microphone is picking up sound.

I pointed Claude Code at that project as a reference, so it would not have to guess.

What we added, one prompt at a time

1. Adjustable zoom and position for the screen recording. I asked Claude to let me adjust and zoom the part of the screen that is being recorded. It worked, and I could pan and zoom while the preview ran.

2. A border highlight. After zooming, it was hard to see where the recorded area ended. I asked for a visible frame around it, and it needed to be preview only, so it helps me but never appears in the video. Claude had already planned for that. I later asked for the border to be blue because it was easier to see on some backgrounds.

3. Better layout. The screen was too small and the face cam sat too low. Over several prompts I asked Claude to expand the screen area to the right, move the face cam up and toward the center, and keep a very slight overlap between the two, like the reference image. Small layout requests work best when you say exactly what should change and what should stay the same.

4. A full face cam mode. Intros are just a talking head, so I asked for an option that makes the face cam fill the whole canvas. This also works with the background blur.

5. An audio waveform. The last feature shows a live waveform, so I know my audio levels are good before I record.

Lessons from the session

  • Give the AI a reference. A screenshot or a working project beats a long description.
  • Ask for one thing at a time. Each prompt changed one behavior, which made problems easy to spot.
  • Review the result between prompts. I kept going back to the preview and adjusting.
  • Use text to speech or voice for long prompts. It is faster than typing when you are explaining a layout.

I also say it directly on stream: the layout is not perfect yet, and the screen still feels a bit small to me. The plan was to record my next video with this tool and fix whatever got in the way.