←back to thread

641 points shortformblog | 1 comments | | HN request time: 0s | source
Show context
chatmasta ◴[] No.42950350[source]
I don’t know about officially sanctioned releases, but I feel like I’ve watched entire movies through YouTube shorts at this point… there’s a really simple grift that rockets to the top of the algorithm and also pushes people into a pipeline for other clips of the movie:

1. Clip a movie scene and crop it for vertical aspect ratio (maybe some AI is used here to choose the focus point of the scene)

2. Add royalty-free background music and possibly other tweaks like mirroring the video

3. Title it something generic that doesn’t acknowledge it’s a movie/show, like “College dropout beats Harvard Law grads to the job” for the scene from Suits (Note: for shorts, the title doesn’t matter if it’s algorithmically chosen to play next… in fact at this point the more relevant title is the optional link to a different short… the real title is barely visible)

4. Do not mention the name of the movie/show in the title or description

There are hundreds of accounts producing these shorts on an industrial scale. It’s easy to see how the automation works and also why it’s successful. It’s clickbait (people want to comment or ask for the title, or correct the title to mention it’s actually from a movie); it’s addicting (it funnels people into watching more clips from the same movie… funny how YouTube knows to do that but not that it’s copyrighted, btw); it’s self-optimizing (if the algorithm doesn’t surface the next short, people go looking for it specifically); and of course, it’s automatable (everything from curation to editing can be automated, and just a sprinkle of AI is apparently enough to obfuscate the automation).

What’s fascinating is that YouTube hasn’t stopped this. The shorts algorithm can obviously detect the similarity between clips from a movie, but the copyright/spam detection algorithm can’t detect the same.

replies(2): >>42951042 #>>42951229 #
atVelocet ◴[] No.42951042[source]
What i never understood:

Why not use some kind of interlacing and randomly sort the lines. The result is a valid video file which could be uploaded to YouTube. Then deinterlace with a browser plugin and the random pattern used to scramble the lines. Same can be applied to the audio.

replies(2): >>42951583 #>>42953445 #
1. chatmasta ◴[] No.42951583[source]
Not sure I’m understanding you, but it sounds like you’re asking why not upload a video that’s scrambled until viewed with a browser plugin that knows how to unscramble it?

That would be cool, but it won’t be very effective as a viral video if everyone needs to have a browser plugin installed :)

The challenge here is to circumvent the copyright algorithms while still looking like a normal video to the user (who has no external tools installed).

However, for things like hosting pirated streams or sharing content out-of-band, it would be interesting. It’s basically the a minimally lossless form of steganography.