Google's New AI: Dog Goes In, Statue Comes Out! 🗽

📝 The paper "DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation" is available here:

Try it out:

AI Image interpolation:

Felícia Zsolnai-Fehér’s works:

Judit Somogyvári’s works:

❤️ Watch these videos in early access on our Patreon page or join us here on YouTube:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Benji Rabhan, Bryan Learn, B Shang, Christian Ahlin, Eric Martel, Geronimo Moralez, Gordon Child, Ivo Galic, Jace O'Brien, Jack Lukic, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Kyle Davis, Lorin Atzberger, Lukas Biewald, Luke Dominique Warner, Matthew Allen Fisher, Michael Albrecht, Michael Tedder, Nevin Spoljaric, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Rajarshi Nigam, Ramsey Elbasheer, Steef, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.

Károly Zsolnai-Fehér's links:

[CorridorCrew]

Pretty funny coincidence that you dropped this video on the same day we dropped our Dreambooth in Stable Diffusion video. I’m curious what you think!

[Terminalss]

Imagine combining this with an AI that turns images into 3d objects. You could take a few picture of your dog, have the ai turn it into a statue, and then 3d print it with no knowledge of modelling software

[MayorMcC]

I remember everyone clowning on the "ENHANCE" trope in CSI shows, maybe they just knew something we didn't

[Rokinso]

Graphic designers: How fast will we lose our jobs?
Google: Yes.

[zenu]

Corridor Crew just uploaded an impressive demonstration of this AI too. Talk about timing!

[CappaBeta]

If you've lost someone, and had been documenting your life and loved ones with good photos etc. Then you could generate novel images or forms of someone who's passed.

In fact, I might go a bit further. You can now start making sure that you have good archive content of yourself, to be preserved. Image, video, speech, chat logs. Your children could talk to an ai version of you, once you're gone.

[richskater]

You could gift an art loving friend a Van Gogh of their pet from the photos they post on social media. What a time to be alive!

[AllisterVinris]

The speed at which all of this is evolving blows my mind.
How much time before we can do a text to 3d model (with good texture, correct material properties, normal map and all that)

[thepositeer]

“What a time to be Alive” I really love this phrase.

[israelRaizer]

So this is basically Textual Inversion from Stable Diffusion but applied to Imagen... Nice

[Nerogar]

There is a technique called textual inversion that is widely used in combination with the Stable Diffusion model. I guess this is basically the same thing?

[Qsie]

This is going to be so cool! I can't wait, my hands are hurting from squeezing these papers 😄

[IAmNumber]

Only a few years ago I thought art would be one of the _last_ tasks to be automated. The fact that AI is already this advanced, this shortly after its widespread adoption means that sufficiently developed machine learning technology will be able to automate far more of the current production process than was previously thought possible.

[Bo-kqtn]

crossing two animals together is going to be AMAZING for creature design!! This is so fantastic. The hippo dog is adorable 😂

[aelolul]

Corridor Crew also released a video today that showcases this technology using themselves. This stuff is mind blowing!

[satibel]

Next paper will probably be about multiple [V] in the same picture, spacial/count awareness or text fidelity.

[redandpigradioshows]

If only Google was the type of company to release to the public, or even go open source

[Lugmillord]

The speed of progress for AI image generation this year is so incredibly fast! Never has there been a science topic where it was this difficult to stay up to date. It's like every week or two there's something new.

[PostmanStudio]

As a CG artist, I don't think this kind of AI will ever replace us. This will give us the opportunity to do a lot of design research at a meager cost before working on the final product. Result in the end, client satisfaction at the same or lower cost.

[TimBlokdijk]

I can see a future where you take a few input pictures to generate 360 degree photo's and then do photogrammetry to create a 3D object?