Two people. One room. One hour with the windows shut. We pointed an Aranet4 CO2 monitor at the problem and watched the air quietly go bad — then fixed it in minutes. Here’s the experiment, and, as always in this series, an honest look at how the video itself was made.
What actually happened
We started in the green — fresh air is under about 500 parts per million (ppm). After an hour of two people sitting, talking and breathing in a closed room, the Aranet4 read 1404 ppm. With a bit more movement and conversation it climbed to 1809 ppm — deep into the red. Then we opened a window and added some airflow. Within a few minutes it fell to 851, and then to 680 — back in the green.
That’s the whole point: the number you can’t see is the one that makes a room feel stuffy and makes you feel foggy. A monitor makes it visible, and ventilation fixes it fast.
What the number means
- Under 800 ppm — fresh, well-ventilated air.
- 800–1400 ppm — getting stuffy; time to open a window.
- Over 1400 ppm — poor air; ventilate now.
For reference, outdoor air sits around 420 ppm — which turns out to be a handy way to check your monitor (more below).
How the Aranet4 knows: NDIR
The Aranet4 uses a technique called NDIR — non-dispersive infrared. It shines infrared light through the air inside the sensor. Carbon dioxide absorbs one specific wavelength of that light, so the more CO2 there is, the less light reaches the detector on the other side. Measure the drop, and you have your ppm reading. It’s simple, stable, and it’s the same core technology in both the pricey monitors and the cheap ones.
How to check yours is honest
You don’t need a lab. Two easy tests:
- Take it outside. Away from your breath, a healthy sensor should settle around 410–450 ppm — close to the global outdoor baseline of ~420. That’s your free accuracy reference.
- The box test. Seal the monitor in a box with a CO2 source — your breath, or a little baking soda and vinegar in a separate cup. The reading should climb quickly, then fall when you open the box. That proves it responds correctly. (It doesn’t prove the exact number — only the outdoor test does that, because you don’t know precisely how much CO2 you added.)
Premium vs budget
We also have a much cheaper monitor — the Tadeto mini — which uses the same NDIR technology for a fraction of the price. Same physics, very different price tag. A proper side-by-side “do they agree?” test is coming in a follow-up.
The Short
How this video was made — and edited
This is part of an ongoing series experimenting with an AI-assisted video workflow, and part of the fun is showing the messy middle, not just the finished cut.
The footage was filmed on an iPhone and the narration is my own voice. The new thing this time: every science graphic — the CO2 traffic-light scale, the NDIR animation, the “over the hour” chart and the “test it yourself” card — was built with Clipkit, a tool that turns a JSON timeline into a rendered video. Instead of hand-drawing overlays, the graphics are described as data and rendered locally, then assembled with ffmpeg.
The first pass wasn’t right, and that’s the interesting part. The initial cut had portrait phone clips dropped into a widescreen frame with blurred bars on the sides, a calm acoustic backing track, small on-screen text, and AI-generated narration standing in for my voice. Watching it back, the notes were clear: the portrait framing looked lazy, the music was too sedate for the subject, the graphics text was too small, and a synthetic voice wasn’t as good as just recording my own.
So the edit: every portrait shot is now punched in to fill the widescreen frame (no bars), centred so the big CO2 number is never cut off; the music is livelier; the graphics fonts are bigger; and the voiceover is my own, recorded in one take and cut to fit. It’s a good example of the workflow finding its feet — the AI does the heavy lifting on assembly and graphics, and the human calls are still what make it watchable.
Total tooling cost for the graphics and render: essentially nothing (Clipkit renders locally, and the only paid step is the voice work).
More experiments at roverplanet.co.uk. If you’ve got a CO2 monitor, take it outside and see what it reads — then come back indoors and watch it climb.
