Every single sample they posted is bad. None of them achieve the intended effect or goal of replacing an experienced photoshop edit.

MIT doesn’t mean everything is news. Another university lab could have produced the same thing with the same results. FastCo wouldn’t have written about it and nobody would have cared.

Yeah. Even some really simple nearest neighbors + blending works way better than any examples in this article

http://graphics.cs.cmu.edu/projects/scene-completion/scene-c...

This seems like it works worse than photoshop’s content aware fill, which tries to do the same thing with a little more user guidance.

Fortunately [0], it doesn't quite work,

[0] https://en.wikipedia.org/wiki/Censorship_of_images_in_the_So...

Previously: https://news.ycombinator.com/item?id=17900750

During some voluntary work, I had erased the ex- of some fellow from an old photo 15 years ago. It was cool, and every body lined up to get a lot of stuff erased. Yet soon it will become trivial!

Deep Stalin would be a more appropriate name.

In other news, they can CONSTRUCT fake videos and photos. Not take away only.

We will need digital signatures to become mainstream for checking sources. Otherwise video evidence for a courtroom or court of public opinion will be very dangerous to rely on!

And “detection” is a no-go because this arms race is like an AlphaGo situation where the difference between the truth and a fake becomes so small that no one will be able to tell.