[mysteria]

To be clear, traitor-tracing each stream would require a unique encode per account to inject the stegotext; that's computationally unfeasible. Doing one encode per movie theater would still be a struggle, but less so by three orders of magnitude.

If the movie is streamed in chunks, only certain short segments would need to be reencoded to add watermark data. Alternatively it might be possible to splice in a short segment with the watermark between keyframes of the preencoded film.

Finally all of this could be done on the audio side which is much less computationally intensive compared to video.

[thr0w]

> If the movie is streamed in chunks, only certain short segments would need to be reencoded to add watermark data

Look into A/B watermarking - https://techdocs.akamai.com/adaptive-media-delivery/docs/add...

[kmeisthax]

If you were only watermarking short sections of the video, wouldn't that make it possible to analyze the stegotext and erase it? You could have a handful of people rip the same video and then compare them, and if different sections get watermarked then you can reassemble an unwatermarked file. This also applies to splicing in short segments of watermarked video.

If you have the whole thing watermarked then all you can do to fix that is averaging; which might not even destroy the stegotext.

Audio watermarking is definitely an option; hell, there's already a DRM scheme called Cinavia that relies on watermarking[0]. If you cam a movie and play it on a Blu-Ray player, it'll actually trip this DRM scheme and, at a minimum, mute the audio or refuse to play the file. I would argue this is probably the most successful use of watermarking, at least in terms of "how much piracy does this frustrate"; but even then you can just play your cams on something else and get around it.

And this is all assuming your CDN provider offers cheap-enough edge compute to inject watermarks before the video hits the user's device. I haven't looked into this recently, but I remember early DRM schemes having very silly bypasses[1] because CDNs could only serve static files. Someone else linked to Akamai documentation about watermarking, but I have no idea how much extra that costs or how much it might complicate other parts of the setup.

[0] https://en.wikipedia.org/wiki/Cinavia

[1] e.g. Remember when someone made an iTunes Music Store client that just didn't encrypt anything, because all the encryption was done on your own device?

[mysteria]

Sectional watermarking is always going to have a higher risk of detection using multiple rips but that's the tradeoff you get with computational power. As you said the best option is to watermark the whole thing but that's expensive.

Cinavia looks interesting as it's done on the client side, like how programs like Photoshop detect the watermarks in banknotes to prevent people from using it to create forgeries. If they managed to get it into the firmware of every television, AVR, etc. then it would be much more effective than just having it on Blu Ray players.