Google Deepmind announces Genie, first generative interactive environment model. .

Gamernyc78

MuscleMod
28 Jun 2022
20,386
16,652

🧞 Genie: Generative Interactive Environments​

Genie Team
We introduce Genie, a foundation world model trained from Internet videos that can generate an endless variety of playable (action-controllable) worlds from synthetic images, photographs, and even sketches.

frN31JMEJARGJkWJiubdOh3UQmi055uvbRfgMVl52VcD8isecJcrJWPoFhH1htONKeS1j7sjGabanvZc68YCumofjOuEhKnKZKgmpmpLiJEN317j9PdyqFXeEKn8A5wGxQ=w1280
Read Paper

A Foundation Model for Playable Worlds

The last few years have seen an emergence of generative AI, with models capable of generating novel and creative content via language, images, and even videos. Today, we introduce a new paradigm for generative AI, generative interactive environments (Genie), whereby interactive, playable environments can be generated from a single image prompt.
Genie can be prompted with images it has never seen before, such as real world photographs or sketches, enabling people to interact with their imagined virtual worlds-–essentially acting as a foundation world model. This is possible despite training without any action labels. Instead, Genie is trained from a large dataset of publicly available Internet videos. We focus on videos of 2D platformer games and robotics but our method is general and should work for any type of domain, and is scalable to ever larger Internet datasets.


Learning to control without action labels​

What makes Genie unique is its ability to learn fine-grained controls exclusively from Internet videos. This is a challenge because Internet videos do not typically have labels regarding which action is being performed, or even which part of the image should be controlled. Remarkably, Genie learns not only which parts of an observation are generally controllable, but also infers diverse latent actions that are consistent across the generated environments. Note here how the same latent actions yield similar behaviors across different prompt images.

_7Kd4DPROhTAF9XxJEdxmn1kE5jZ-2kWrQO_oTiqT4U7osRVhdScmc_93D1f8a0bO-xnaLHlUkATWZUmhkpwCjWfZJjAASzTOH8i5oxqVwMSWX6QCWOYtnJ7mMeS3ZZCYA=w1280

NeRQfGKuO0cGMvL6SxaKfumOjURzkK0_RNBSjf1AP_5_ZaXoxNpxiR8Bxm7Qfoxyv3tpkyfZepBG4XQAVTnX4hN3W2KUJL0P8idyO-WByp3jVOGK9935dJPNZrgpkjNRIA=w1280

GcuaWcgJXO9Dmu3_QNxpHcCYOIJihTbS6lhiR3gcVbPYV0HmGcJ8uYeJrBoV_sg7UH-ZLIvmsDmU-eTvNa7-R6ULEkAulRfTbg3bXgG2jkm10G4-uDbwaSLPJbh1_gUoIQ=w1280

Enabling a new generation of creators​


Amazingly, it only takes a single image to create an entire new interactive environment. This opens the door to a variety of new ways to generate and step into virtual worlds, for instance, we can take a state-of-the-art text-to-image generation model and use it to produce starting frames that we can then bring to life with Genie. Here we generate images with Imagen2 and bring them to life with Genie.

HQMCV-mVzsaO_8qzEWbmjTxIPAZawO621EoYcTV9rnv55b2u91wxak1BksL85UgZsIp1cGhJKDn5XKMKgubpr2Xx3R37Nzif5w0EULhZ91Rqw4UEvFgD9qM_OpXcDgPNpw=w1280

nSTGFq2-32XMXN1JJMZEYeY0D1RQ2ZEmJDtfYndClv4YKgHEaEY3hVhiHKCaVzlO2Fg942PEZQM1SrgfpLlxzWbIWe6pwbmLLMUPqt492zETJFAZxkpvAkF5j6ZoYxsiiQ=w1280

aNGBsL-qvTZoXJCeGlPxCHrOsiXoEyRZxd1YWA8cY-6yM643_VjDMC32qWQ7VPDCasQxtFmfywA2Loaxh6AAEoZPdJigJruRsEJy5dL-lnMLtD-7ErcxVyJI4FeEMe1Hdw=w1280

kmSXtjqySaINacF3KX8rkGAx_YbhVLKR7l6982Sx6iqTwabvrXKqGLIVa3DosPNQoN70Ur3r7VCUyUGAwEn3pGsSNYdNlPWy-ekhlQNRNZrksICJ8UF2-iCsIhAR5wxmsw=w1280


piENqtZJuZU0-RLxGyqqGwoc7RGIcfRt7jQDwlIMvMuCrHfA3TfqNKrXqS7W92xSepnQj3Kx25LwH-bP9HoEBwwKVS99dRLmoyUiECkRMUhDBoikoEibO0iFWSnw65QJgg=w1280

xH0s2K-YnNW7iz54lFRA0QqPwCxTJ8jDRhVDjGkQEsFmPjOkh6ZD2gXozDLFCZLX2gD33suP8ySd2l5jt8Va0H_tbunLGtysudumKwfR2UXSLP5E_6nwQsacHU4AvZ2APw=w1280

28j1t3huyg9g-l8sepxoNBTUuB0Zx5nIBliz0NfCNhhoMs-Dg-2vgIOG2r1O-ZLcI8lyphwpngdJv4HtcRmOSjxuebQUEMOn4g35PcQ4BxTkuLbmpMWd0ym372LtL-s_bQ=w1280

4tgUQSTT7ISkRg7Yjva8FouyPaP06BgFfQ1LJ8wyACkUFzApOLjmdcApYPZEklYyGLDdsPXopSK0IfdDPoEAEe5YdL1PEhTSIUbUqoVcpiyJwLkcIq8pwRfehrz1w53jYQ=w1280

But it doesn’t stop there, we can even step into human designed creations such as sketches! 🧑‍🎨
 

Box

May contain Snake
6 Apr 2023
3,500
3,759
More AI Slop, ubisoft and Microsoft is going to have a field day...
 
  • sad
Reactions: JAHGamer

Nhomnhom

Banned
25 Mar 2023
8,414
11,558
More shovelware coming up. It's already used to make tons of gaming articles tht often times are incomprehensible nonsense lol
AI pretty much already ruined the internet as it is. Whenever you look anything up on google now you get inundated with nonsensical articles. Only moderated stuff like reddit will actually end up having the information you are looking for.
 

Box

May contain Snake
6 Apr 2023
3,500
3,759
AI pretty much already ruined the internet as it is. Whenever you look anything up on google now you get inundated with nonsensical articles. Only moderated stuff like reddit will actually end up having the information you are looking for.

Art pages on instagram are ruined, AI "art" slop is now cluttering up the whole feed.

We already have a problem with indie shovelware with video games, now with AI the problem will be much worse
 
OP
OP
Gamernyc78

Gamernyc78

MuscleMod
28 Jun 2022
20,386
16,652
Art pages on instagram are ruined, AI "art" slop is now cluttering up the whole feed.

We already have a problem with indie shovelware with video games, now with AI the problem will be much worse
Yeah too much I curated garbage.
 

Airbus

Veteran
30 Jun 2022
2,447
2,162
No, it's just good old fashioned greed powering these layoffs.

Besides, if AI could make a fully functional AAA game, it'd just be making very bad Assassin's Creed and Destiny clones, because that's the type of low effort slop prompters could muster.
Not necesarilly taking over everything ( yet)

But with AI capabilites and constant improvement in creating motion capture, etc that surely will get rid of the needs for many talent artist in videogames



 

Airbus

Veteran
30 Jun 2022
2,447
2,162
Might as well become a retro gamer if our games end up becoming AI hallucinated nonsense.


My money is on Totoki and the other bean counters at Sony selecting someone that is somehow worse than Jim Ryan but I wish I was wrong. The next PlayStation CEO should be someone like Ted Price.
The bottom video explain everything

Videogame animation that took weeks or month AI can do it in just couple hours

Why hire and pay people when AI can do it for free and much faster
conway wow GIF by University of Central Arkansas
 

Nhomnhom

Banned
25 Mar 2023
8,414
11,558
The bottom video explain everything

Videogame animation that took weeks or month AI can do it in just couple hours

Why hire and pay people when AI can do it for free and much faster
conway wow GIF by University of Central Arkansas
Why make good and profitable games when you can make AI junk with lower costs like all the other trash publishers chasing the next trend?

It's like they are unable to learn from their own mistakes and keep making unnecessary changes to what is already working better than ever (due to decisions made and a development culture established more than 15 years ago.

The Last of Us, Uncharted, God of War, Gran Turismo, Death Stranding, Bloodborne, Spider-man, etc, didn't come from doing the least amount of effort hoping to save costs and increase margins.
 
Last edited:
  • Like
Reactions: Gamernyc78

Yurinka

Veteran
VIP
21 Jun 2022
7,719
6,605
Videogame animation that took weeks or month AI can do it in just couple hours

Why hire and pay people when AI can do it for free and much faster
Not true. First, to make an animation doesn't take weeks or months.

And second, has happens with drawing or music or code, AI can't generate exactly what an artist or coder has in his mind. But it can generate a somewhat close and not viable stuff for final work result useful to brainstorm and as a draft or point to start to work on top, helping to speed up their job.

AI just put Rockstar talent artist hardwork to shame


And this is just the begining
No, what is done there is giving screenshots of Rockstar games and asking the AI to slightly modify it to make them more realistic.

Doing it by hand, the artists achieved exactly what they wanted. If didn't have use the trailer screenshots as reference the AI wouldn't have shown such similarity in all the details.

Also, to make that process of making a game screenshot look more realistic and detailed (can easily be done wih somelike like Stable Diffusion XL) nowadays takes some seconds. It can't be done 30 or 60 times per second (on top of the work required by the console/PC to render the game and handle the gameplay), can't be applied in real time for gameplay.
 
Last edited:
  • brain
Reactions: The Icon