So there is no interactivity, but the generated content is not the exact view in... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		dabochen on Aug 28, 2024 \| parent \| context \| favorite \| on: Diffusion models are real-time game engines So there is no interactivity, but the generated content is not the exact view in the training data, is this the correct understanding? If so, is it more like imagination/hallucination rather than rendering?

famouswaffles on Aug 28, 2024 [–]

It's conditioned on previous frames AND player actions so it's interactive.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact