How does one determine what is or is not a deep fake? Is this a game of cat and mouse? How does one handle false-positive mislabeling? Why stop at images?
The identification of deepfakes is a current and hot research sector in computer vision, and a cold war between advances in deepfake technology and those seeking to identify artifacts of deepfakes. It's currently far from trivial to simply add a filter. Google Arxiv to find out more:
Not all the generated is with the sophisticated software. And they leave some trail of their source in their exif information. And images generated with popular software can be identified too.
Is a label like this feasible? Many people trust Google as authentic source, I remember a non-technical me can easily be fooled with a morphed image.
[1] https://arxiv.org/pdf/1911.00686.pdf [2] https://ai.facebook.com/blog/reverse-engineering-generative-...