Still using invisible watermark package that rarely works?
I'm surprised version 2 is still using the https://github.com/ShieldMnt/invisible-watermark package which rarely works. The encoding scheme is just not at all robust. The github itself has years old issues submissions of people complaining that it doesn't work that remain unaddressed and unresolved.
Either excise the package completely, since it is almost completely useless, or write your own watermarking code. Maybe something more robust like hamming codes + pixel rounding? (Or you could at least fix the above package so it solves recursively to force it to properly decode into the watermark).
Or just comment it out of your code and move on
Or just comment it out of your code and move on
The code runs fine with the line in. It just doesn't work.
Have you considered that the correct watermarking of AI images is perhaps actually both useful and important? For the future development of AI image generators and for wider society?
It is a great idea to have them watermarked. But it's shocking to me that no due diligence has been done to check that it actually works.
Edit: apologies, accidentally misclicked on my phone and closed the issue temporarily.