Nopeify

Drop in a photo and it comes back with an AI watermark in the bottom right corner. That is the whole tool. Everything runs in your browser, no uploads, no servers.

Try it on a real photograph of your own, then look at the result and notice what your eye does with it.

But why?

A watermark in the corner now does a lot of arguing on its own. Enough people have learned to read one as a sign that the image is generated, and once they do, they stop looking at the image itself. Nopeify takes about forty lines of canvas code. That is the entire cost of taking a real photograph and making people doubt it.

The uncomfortable half is the other direction. A generated image does not have to carry a mark at all. Watermarks are applied by the tools that choose to apply them, they are trivial to crop, and nothing stops a model from producing an image with nothing in the corner. So the mark tells you almost nothing either way. Its presence is not evidence that an image is fake, and its absence is not evidence that an image is real.

That gap has a name. Brookings calls it the liar's dividend: once convincing fakes are known to be possible, real footage can be waved away as fake by anyone who finds it inconvenient. The forgery is the obvious harm. The deniability is the one that lasts. I collected that report and the rest of the reading on Saine.