There are a lot of weird models out there that are doing increasingly obscure yet interesting tasks

Today, I want to introduce you to Netflix’s Void Model, which removes not only objects from videos but everything that the object has an effect on.

Imagine you had a video of dominoes falling down, then you just removed a few of the dominoes in the middle.

Once this model runs through the video, the first domino falls flat, and the dominoes after the missing dominoes just stand there.

In their examples, they have another example of 2 cards plowing into each other in the source video, then the video the model pushed out which removed the second car, and the first car just kept cruising along.

You can tell the model what you want removed by passing in a quadmask, which is basically where you paint over what you want removed frame by frame, and the model uses that to determine what needs to be removed.

I know I am skeptical about AI uses, but this one is a fairly mind-blowing application.

AI has gone from barely generating Will Smith eating pasta to being able to have such an in-depth understanding of world objects that you can just tell it to remove an object from a scene, and it can infer enough about the physics and the interaction that object would have on that scene that it can render something halfway decent.

It’s not teleportation or time travel, but impressive nonetheless.

While impressive, it is equally scary, as I am sure this could be used to rewrite history for nefarious purposes.

As they say, “Believe nothing you hear, only one half that you see, and nothing that was run through this model”(jk).

I want to thank my Discord mods for sharing this and a lot of other interesting stuff with me. If you want to get in on conversations like this and all the other stuff I talk about, you should jump on my Discord.