DeepMind’s New AI Found The Sound Of Pixels!

If you are interested in a sponsorship:

DeepMind's Veo:

Gen-3 text to video (note: the new gen-3 requires a subscription):

Video sources:

📝 My paper on simulations that look almost like reality is available for free here:

Or this is the orig. Nature Physics link with clickable citations:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Kyle Davis, Loyal Alchemist, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.

[pyropimp]

Waiting patiently for the new wave of "Will Smith eating spaghetti" to drop.

[gabe]

In recent years, I have prided myself on my ability to recognize AI images and sounds, but this is almost unrecognizable from reality. I am both amazed, and terrified.

[SHPfz]

the fact that it even knows when the car drifts

[albert]

The smoke text part is one of the most surprising part! Unintentional skill!

[Estrallio]

Exponential growth, I am scared and fascinated

[catakuri]

What i'm really waiting for is to be able to play a game where you're inside of an AI's dream, so the AI has to generate a 3D environment and update it in real-time as the player moves/interacts with things

[DefaultFlame]

The people saying that AI is "over" need to get their heads checked. Where AI is now is where computers were in the 70s or 80s. We are still just scraping the surface, and every single year results in an unbelievable leap forward from where we were.

[FancyFeast]

As a video professional who's been studying AI for the past few years, SORA and other video AI's caught me by surprise. They must have been pouring R&D into them quietly for a while. So impressive.

[NoahHornberger]

particularly interesting, is the way the techno music was also linked to the car in the city scene. this is of course due to the many videos that already have such sounds in them in that theme. So the question is, will our past style choices continue to inform what an AI considers appropriate and will it become good at doing what we would expect rather than what would be more creative?

[MrtnX]

Dear scholars, AI is holding on to our pappers now.

[sequenc-]

Next year, ai generated media will be virtually indistinguishable from real life, from the image, sound, motion, and story telling. This is actually scary

[richcolour]

5:13 dog walks like a cat and has an ear deficit

[JackC.Johnson]

your video style is incredibly unique and refreshing! ️

[bushwentto]

0:08 "Text prompt goes here" lol

[TheBlueNeutron]

I thought it would be atleast another two years before anything like this. I am simply dumbfounded.

[rangerCG]

There is time/motion/video data in all photographs with motion blur. It would be cool if an AI was trained to pull an approximation of motion/video out of really old photographs, which tend to have a lot of motion blur due to the long exposure times needed.

[xenontesla]

Very interesting! It’s kind of odd that AI generated videos were used for the audio samples, especially the drums and guitar. It would’ve been nice to have real video with audio to compare it to, and the video generator isn’t quite generating hand movements that make sense.

[lukasgruber]

lets be realistic, 90% of the mass usage will be in the "adult department"

[corediagram]

This is actually real-time growth and deformation via videos as an early iteration. The key is noise outside the object in the environment which can be manipulated to give a plethora of effects like gravity, wind, viscosity, light effects, sound, etc. Cool video.

[albert]

An insane breakthrough! Synthetic data is going to be crazy! Finding ways to keep track of these synthetic data and whats real is gonna be a trouble in the future!