This is mind-blowingly impressive! Before you know it we will be able to create our own VR games. Our own VR environments. The future is looking awesome 👍👍👍👍
At the core, this is diffusion but with the depth map. So the generated scene should be a cloud of points, not organised meshes with bones and animations. While it's an upgrade from classic diffusion, it still implies strong limits for now.
exactly, i got a few videos using depth maps and blender to achieve this result. Their results are arguably better though, but claiming whole 3d worlds is very disingenuous .
And now we need to adapt it for gaussian splatting and train the models on petabytes of real point cloud data to have realistic scenes like of the time square 😅
I don’t make videos but if I did I think I’d use this over video generation. I could capture the video myself as I walk through the scene and tweak it exactly how I want without needing to regenerate and risk a bad output.
Realtime, on the fly generation of game environments (and even NPCs) that change based on user actions will be amazing. No two games on the same map will be the same. For that matter, the AI will create entirely novel maps. Replayability will be essentially infinite.
They need to add the ability to add effects and physics to single objects next, like picking up and moving rocks or walking up to a random table and removing the salt shaker. Later they can work on people. God image being a game programmer or bug checker and turning an image into a 3d space and then having to check every object to decide what's moveable or not. Craft routes for moving objects like people and cars, ad triggers for dialogue and events.
Great episode thanks. 🙏👍 As an aside, UA-cam could implement a new feature x4, x8, x16 playback by cloning the presenter's voice and rewording to cut out the yada yada.
If AI can be trained to generate entire 3D worlds from a single image with this level of precision and creativity, imagine the potential for creating the ultimate game engine. By training AI specifically for graphics and physics simulations, we could achieve hyper-realistic visuals in 4K at 60 FPS effortlessly-far beyond what manually written code could accomplish. This isn't just innovation; it's the gateway to redefining gaming and interactive media. Just saying... you should totally explore this idea!
If this is what I think it is then it's not a 3D world, it's a 2.5D set, which is why you wont see the camera moving very far. Pretty useless for game development outside of niche scenarios.
if they have gotten there AI to work on geometry rather than pixels it means it can work at more than a picture by picture level. it will be one of the very few western AI to get past that barrier almost all of them are limited to a single picture cant predict a video ecxept to guess image by image with greater and greater error range the further u go
Quick summary: World Labs' new AI generates consistent, explorable 3D worlds from single images, a leap beyond 2D AI image generation, impacting gaming, film, architecture, and VR. The system understands 3D geometry and spatial relations, building complete environments beyond the image's visible area, maintaining style consistency unlike typical AI. Accessible via web browsers, users control a dynamic camera for exploration and can apply cinematic effects like depth of field and dolly zoom, enriching content creation. This AI creates true 3D scenes, ensuring viewpoint consistency and adhering to physical laws for realistic visuals, unlike pixel-based AIs. Users can dynamically modify lighting, geometry, and add objects, offering unprecedented creative control and enabling interactive effects within the 3D environment.
It is time to work on casual pair extraction, particularly in the context of emotion-cause pair extraction. We have to reduce false positives, so I can have my fun. BTW! Any human that thinks their going to fake EQ with neuroscience. Oh disclaimer wherefore art thou!
Being able to prompt worlds while being in VR will be pure magic...
yessssssss
This is mind-blowingly impressive! Before you know it we will be able to create our own VR games. Our own VR environments. The future is looking awesome 👍👍👍👍
At the core, this is diffusion but with the depth map. So the generated scene should be a cloud of points, not organised meshes with bones and animations. While it's an upgrade from classic diffusion, it still implies strong limits for now.
Probably why it looks a bit stretched and unsettling
exactly, i got a few videos using depth maps and blender to achieve this result. Their results are arguably better though, but claiming whole 3d worlds is very disingenuous .
@@tyski9000 arguably 3d worlds themselves are indigenous clouds of points mapped into fake cameras and lights
And now we need to adapt it for gaussian splatting and train the models on petabytes of real point cloud data to have realistic scenes like of the time square 😅
I don’t make videos but if I did I think I’d use this over video generation. I could capture the video myself as I walk through the scene and tweak it exactly how I want without needing to regenerate and risk a bad output.
Realtime, on the fly generation of game environments (and even NPCs) that change based on user actions will be amazing. No two games on the same map will be the same. For that matter, the AI will create entirely novel maps. Replayability will be essentially infinite.
Mind blown by the level of detail and realism in these generated environments! Can't wait to see where this tech goes next.
Amazing 🎉
This is definitely the future. So cool.
I am old and not a gamer but the craftsmanship of AI is intriguing.
Imagine going into different isekai worlds with this, with random AI waifus, with VR.
Nahh get out the basement
They need to add the ability to add effects and physics to single objects next, like picking up and moving rocks or walking up to a random table and removing the salt shaker. Later they can work on people. God image being a game programmer or bug checker and turning an image into a 3d space and then having to check every object to decide what's moveable or not. Craft routes for moving objects like people and cars, ad triggers for dialogue and events.
I like this approach a lot more than what we have in the AI world right now. I hope they will have this AI out soon. Can't wait!
Great episode thanks. 🙏👍
As an aside, UA-cam could implement a new feature x4, x8, x16 playback by cloning the presenter's voice and rewording to cut out the yada yada.
Imagine using a photo of your home you grew up in as a child. It would be really bizarre .
Are you stevenmarkryan?
Finally something interesting in the image gen space
can't believe something like this is possible
If AI can be trained to generate entire 3D worlds from a single image with this level of precision and creativity, imagine the potential for creating the ultimate game engine. By training AI specifically for graphics and physics simulations, we could achieve hyper-realistic visuals in 4K at 60 FPS effortlessly-far beyond what manually written code could accomplish. This isn't just innovation; it's the gateway to redefining gaming and interactive media. Just saying... you should totally explore this idea!
If this is what I think it is then it's not a 3D world, it's a 2.5D set, which is why you wont see the camera moving very far. Pretty useless for game development outside of niche scenarios.
3D Worlds!!!! HERE WE GO!!! 🥳🎉... We need 3d modeling for consistency though not just image generation
It reminds me of the Classic series of Point and click games Myst,
PAINTING : Edward Hopper "Nighthawks"
if they have gotten there AI to work on geometry rather than pixels it means it can work at more than a picture by picture level. it will be one of the very few western AI to get past that barrier almost all of them are limited to a single picture cant predict a video ecxept to guess image by image with greater and greater error range the further u go
I could see this with interior design or say, landscaping, etc. Architectural or whatever.
Ai grid ? Is that you or an ai voice ?
Now this is truly awesome. 👏 Mindblowing tech!!
Quick summary:
World Labs' new AI generates consistent, explorable 3D worlds from single images, a leap beyond 2D AI image generation, impacting gaming, film, architecture, and VR.
The system understands 3D geometry and spatial relations, building complete environments beyond the image's visible area, maintaining style consistency unlike typical AI.
Accessible via web browsers, users control a dynamic camera for exploration and can apply cinematic effects like depth of field and dolly zoom, enriching content creation.
This AI creates true 3D scenes, ensuring viewpoint consistency and adhering to physical laws for realistic visuals, unlike pixel-based AIs.
Users can dynamically modify lighting, geometry, and add objects, offering unprecedented creative control and enabling interactive effects within the 3D environment.
Ah great idea, UA-cam clearly didn't have enough bots already posting AI slop everywhere.
She raised the money and got the team in a flash
It is time to work on casual pair extraction, particularly in the context of emotion-cause pair extraction.
We have to reduce false positives, so I can have my fun.
BTW! Any human that thinks their going to fake EQ with neuroscience.
Oh disclaimer wherefore art thou!
If you want to comment first, follow the subscription tab.
Edit: I am second By the way.
So in essence they took unreal engine as a front end generator and feed it AI from the backend ?
Wonder if it's too late to learn Blender...
That's pretty amazing
prompt: let there be light
seems pretty basic Why don't the basketballs cast shadows can't they post render shadow maps onto the geometry?
Inside out gaussian splat
AI seems to be a technology looking for a problem
❤
❤🙌❤🙏❤🙌❤