Skip to content

Comment on Atlas: A World Model for Spatial Intelligenceparent

Comments

Unfortunately that's a complex question... this depends on the number of diffusion steps, the size of the context, the image resolution, and the type and number of inference devices we use. There are lots of knobs to trade off speed, quality, latency, throughput, and cost.

An ideal workflow would be something semi-interactive that you can use to quickly iterate on an idea, followed by a longer offline bake-out to generate final production-quality assets.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.