3 ms·

>Might be a realistic way for manufacturers to to implement a certified-taken-by-camera-not-AI photo feature. How would that work? I would imagine that any sys

>Might be a realistic way for manufacturers to to implement a certified-taken-by-camera-not-AI photo feature. How would that work? I would imagine that any system to implement this would necessarily be something that AI tools could replicate, wouldn’t it?

maronato 2y ago
Using encryption. When you take a picture, the device or app creates a signature using the photo data and metadata. Then you can check the signature using the company’s public keys. If you make edits to it, the editing app will package the new metadata, edited photo data, the original signature, and sign it again. Now you have a chain of “changes” and can inspect and validate its history. It works for video and audio too. As long as the private keys aren’t leaked, there’ll be no way to fabricate the signatures. https://c2pa.org/ https://c2pa.org/
- goblin89 2y ago
  This has existed for a while and it does not require licensing your footage to camera maker.
- 986aignan 2y ago
  Couldn't you replace the CCD with an adapter, connect the adapter to the video out of a computer, and then use the camera to "take a picture" of your already edited picture? It seems to me that any "paper trail" scheme of the sort you describe would have to solve the problems of DRM to work: making the elements that report on the real world (in this case, the CCD) tamper-proof, making the encryption key impossible to extract, designing robust watermarks to avoid analog holes, etc.
  - maronato 2y ago
    Sure, you can also take a picture of the screen. I don’t think C2PA’s goal is to completely prevent this type of thing, but to make it hard enough to stop low-effort attempts. This, like DRM, will probably be an arms race, and future solutions will look nothing like what I described. But then again, the spec has been out for more than a year, and I haven’t seen anyone big bothering to implement it. Maybe it’s a flop already.
- codedokode 2y ago
  An ordinary person might be not able to fool this technology but I am sure 3-letter agencies can easily sign any picture.

maronato 2y ago

Using encryption. When you take a picture, the device or app creates a signature using the photo data and metadata. Then you can check the signature using the company’s public keys. If you make edits to it, the editing app will package the new metadata, edited photo data, the original signature, and sign it again. Now you have a chain of “changes” and can inspect and validate its history. It works for video and audio too. As long as the private keys aren’t leaked, there’ll be no way to fabricate the signatures. https://c2pa.org/ https://c2pa.org/

goblin89 2y ago
This has existed for a while and it does not require licensing your footage to camera maker.
986aignan 2y ago
Couldn't you replace the CCD with an adapter, connect the adapter to the video out of a computer, and then use the camera to "take a picture" of your already edited picture? It seems to me that any "paper trail" scheme of the sort you describe would have to solve the problems of DRM to work: making the elements that report on the real world (in this case, the CCD) tamper-proof, making the encryption key impossible to extract, designing robust watermarks to avoid analog holes, etc.
- maronato 2y ago
  Sure, you can also take a picture of the screen. I don’t think C2PA’s goal is to completely prevent this type of thing, but to make it hard enough to stop low-effort attempts. This, like DRM, will probably be an arms race, and future solutions will look nothing like what I described. But then again, the spec has been out for more than a year, and I haven’t seen anyone big bothering to implement it. Maybe it’s a flop already.
codedokode 2y ago
An ordinary person might be not able to fool this technology but I am sure 3-letter agencies can easily sign any picture.