I think this could extend to major public figures, but offers no recourse to the average joe who doesn't have 100 hours of talking footage online. Deepfakes require a lot of data to produce. Anyone who doesn't have this much data, you could reliably assume the content is real.

"Deepfakes require a lot of data to produce."

I wonder if this is going to hold, especially as 4K video is becoming the norm.

It won't, as with everything in DL the models are becoming much better at training on a large amount of data and fine-tuning on very little. I am pretty sure in 1-2 years we will have semi-convincing one-shot learning deepfakes models with 5-10 pictures you could get on the person's Facebook profile.

"Deepfakes require a lot of data to produce."

Are you sure? https://www.myheritage.com/deep-nostalgia

The result is not perfect but it could be with a few more photos.

And realistic synthetic voice generation is not so far away in the future.

> 100 hours of talking video

You mean like zoom recordings? Yeah nobody has those.

Haha, good point. I didn't even really think of that.