Some or all images on openAI will be invisibly watermarked.
It will not be trivial to remove (by resizing/changing format), even if I can download the image in png or jpg or whatever.
This situation continues for at least 3 months.
The market can resolve YES early if we get 3 months with an unbreakable watermark.
Related questions
Apparently they also have a watermark tool for text, but they don't want to release it because it will decrease usage (all the students that are cheating on their assignments 😂). Source here: https://www.wsj.com/tech/ai/openai-tool-chatgpt-cheating-writing-135b755a?st=3klcje9e80bnevy&reflink=desktopwebshare_permalink
Market for release of text-based watermark:
@RemNi looks like they're only putting in metadata for now and plan to add image watermarks later
Does it count if it requires a specific tool to remove, rather than just image conversion? For a more extreme example, if I add per-pixel RGB noise to remove it? Does cropping count as resizing?
Once it's widely known and testable, there's very little you can do to prevent people from just removing the features that make it visible.
There are steganography techniques that are resilient against image resizing and compression artifacts, so long as you don't want to store much data. A 1 to 16 bit flag saying "this was an OpenAI model" and maybe the specific model, would not be particularly difficult to add. The main business would be decent error correction across a wide range of frequencies. If I was implementing it, I'd encode it in phase information in spectral analysis of the image.
In my view, this is 100% YES if they want one, depending on the strictness of your "triviality".