Cropping 16:9 to 9:16 throws away two thirds of your frame
Clipping a stream into vertical looks free. It costs you most of the picture and all of your resolution headroom, and the maths is not close.
You streamed 1920 by 1080. Your clip tool crops it to 9:16 for the short-form platforms.
Do the arithmetic on that crop.
The maths
Keep the full 1080 height. A 9:16 frame at that height is 1080 times 9 divided by 16, which is 607 pixels wide.
You started with 1920 across. You kept 607. That is 32% of the width and 32% of the pixels.
Then the platform wants 1080 by 1920. Your 607-wide crop gets upscaled by 1.8 to fill it. You are inventing two thirds of the pixels you deliver.
You are not showing 1080p vertical. You are showing a 607-pixel-wide strip stretched to 1080, made from footage a codec had already compressed for a wide frame.
Framing, which is the worse problem
Horizontal framing puts information at the edges on purpose. A guest on the left, a game on the right, an overlay in a corner. The centre is often the least interesting part of the shot.
A centre crop takes the boring third and discards the reason the shot was composed. Auto-reframe tools track a face and pan, which fixes the subject and adds motion nobody asked for.
What actually works
| Approach | Cost | Result |
|---|---|---|
| Centre crop from 1080p | Free | 607 px of real detail, upscaled. Soft. |
| Crop from a 4K source | Storage and encoder load | 1215 px of real detail. Holds up. |
| Second vertical scene in OBS | Setup time, one canvas per output | Composed for the format. Best result. |
| Second camera shooting vertical | Hardware | Native. Also the most work. |
The cheap version that is worth doing
Record to disk at 4K even when you stream at 1080p. Storage is the cheapest thing in the rig.
Crop your verticals from the 4K file. Now the crop is 1215 pixels wide and you are downscaling to 1080 instead of stretching to it. Same workflow, and the clips stop looking like clips.
Sources
The Rig Wire reports on streaming technology from public documentation, source repositories and vendor releases. We did not bench-test what is described here. When we have tested something, we say so.