@yellowbeard@queenlainsrevenge.xyz >and to automate it would have looked so bad that it would have captured noticeable chunks of the pavement and grass. Yeah, I suppose it has gotten better but it's still pretty janky. Used to be like that. >But his body is also there, and it's behind the mics. Correct. It's obvious to us but this is a whole field of computer science. Not necessarily obvious to an algorithm doing it. >That would have to be different layers. Yes but it's not like a person is creating all the layers. >If it thought the mics were background, it would put them behind his body too I seriously doubt they're using a neural network to do this. It's just an algorithm. It doesn't know that. Image reconstruction is hard yo. >because the contrast between his body and the zero alpha channel around it would be such a high contrast and would stand out so much that you would basically have to go in and edit things by hand to put his hand over the mics with his body behind them. Do you not see how fucking different in contrast his hand is from his suit? His hands are also moving differently relative to his body whereas the mics are just single solid entities moving. Likely thought the hands were a different entity from the body. Could tell the sleeves were connected to the hands but likely didn't know the sleeves connected the body and it was a single entity. Image/pattern reconstruction is hard yo. If you think that's dumb and it'd be easy to make software see "obviously this is a person so it's a single layer" then I look forward to paying for your software because you'll have some of the best in the industry and be on the cutting edge of pattern reconstruction. You could easily become a billionaire. That is if it is indeed as easy as you seem to think it is.