a lot of the big names in image generation fingerprint the images they output to make them easy to filter out of training data. theres a possibility they could detect this fairly easily.
Typically, I think that requires the photo to be modified for the purpose of identifying it later. I don’t think any old photo can be retrofitted with such a capability posthoc.
a lot of the big names in image generation fingerprint the images they output to make them easy to filter out of training data. theres a possibility they could detect this fairly easily.
Is that the sort of thing that would survive screenshotting and messing with the color balance?
Or even taking a picture of the screen using a phone
I’m not familiar with the exact techniques used but I am aware that there are ways of fingerprinting images that would survive such transformations.
Typically, I think that requires the photo to be modified for the purpose of identifying it later. I don’t think any old photo can be retrofitted with such a capability posthoc.