“The future of video models will converge towards world models soon, where we will eventually all want to skip direct image/video output, and l generate 3D scene / simulation data on demand so we can render at maximum speed + control for real time / immersive / spatial. Internally these latest video models generate a 3d scene graph in latent space, which I noted in my 2024 GTC talk. We are now even fewer steps away from extracting and harnessing this internal data for use in neural rendering workflows that combine all the benefits of traditional 3D rendering with the power and artistic augmentation of Generative AI.” @JulesUrbach 27.02.25 $RENDER