Nvidia clarifies DLSS 5 infers on 2D screencaps plus motion vectors and does not use existing geometry, textures or lighting

Nvidia confirms that DLSS5 is just taking a screenshot of the display output and is analyzing it with a generative AI model in order to “enhance” the visual fidelity. The AI model essentially decides

what it thinks the final render should look like with little control over the output from the artists. ...