PixelUMM
by NVIDIA · Model · Image
mixed
Interesting research preview that generates images and video, but below frontier quality
Nvidia unified model that understands and generates images and video directly in pixel space, with no latent encoder or decoder
StatusResearchAccessOpen weightsLicenseApache-2.0
Why we say so
NVIDIA describes a single model that understands and generates images and video directly in pixels, with no latent encoder or decoder. AI Search, who tried it, found quality below frontier models and called it a research preview. The code and model are out, which makes it useful for study more than production.
Evidence
- Oct 4Quality below frontier models, a research preview; code and model released, mostly Apache 2 licensed