Is that not the same as the second? The second method can be generally be detected programatically, but is invisible to the human eye. I was going to question

Is that not the same as the second? The second method can be generally be detected programatically, but is invisible to the human eye. I was going to question the usefulness of an invisible watermark, as you could otherwise just compare a hash of the image, but I suppose an invisible watermark would likely persist through re-editing / resizing?