I accidentally recorded some vids of beam shots in vertical mode. I have since confirmed that there is actually no way to correct such a problem without destroying the image quality. There are hacks and tricks to offset it like blurring the background and creating weird image tricks, but they aren’t worth it at all.
I totally ruined - or really, the phone ruined by not adjusting quick enough - 4 good beams hot videos. This has to anger the most sagely among us!
We are dealing with goddamn AI and yet we cannot fix an orientation!
My wife is always making videos in portrait mode. Then wonders why when she tries to stream them to the TV they are still in that mode and won’t take up the whole screen. I keep telling her to change the orientation before she records. My advice falls on deaf ears. As does just about everything I suggest. The married life I reckon.
In the first place smartphones could record the actual full size of the sensor (4:3) instead of cropping it to 16:9, maybe save both original and cropped videos. A 3:4 video is certainly better than 9:16.
Or they could even use square sensors, then it wouldn’t matter in which orientation the video was taken.
Edit : there is a 1:1 option on my smartphone, but the original 3:4 is cropped to 9:16 and then to 1:1, it’s not possible to record the full sensor, absolutely ridiculous.
You’re right, that’s an interesting thought! I was just considering this the other day too. Perhaps a future AI tool from OpenAI could fill in the missing horizontal content frame by frame, similar to how DALL-E works.
Take a 10-second video everyday of your shin. Every time you screw up and do it vertical rather than horizontal you have to hit yourself in the shin with a 2x4. Even if you’re a slow learner the problem will be solved within a week. And you have to show us the bruises.
Indeed, there is still no way for a computer to accurately fill in missing data that it doesn’t have. Computers are getting better at guessing what might have been past the edge of the data, but it will only ever be that – a guess.
If you want the data to exist and be accurate, the only fix is to record it correctly.
So this isn’t exactly what you’re talking about , but I do think it’s relevant to the conversation. Have you seen DALL-E? It uses AI to extend images to the sides. Two caveats:
The extension on the sides of the original image isn’t real.
I would very much like to be able to preselect a square shape for the images and videos my phone produces. I had a Samsung Galaxy once that had a square image as one of the available formats that could be selected. For some things, I used that setting a lot. Some old roll film cameras produced square images.
Almost all camera apps allow you to set 1:1 (square) for photos. I use that a lot for creative photography. You will give up a bit of your sensors max resolution for this, though.