AMD Nitro-P: Pixel-Space Text-to-Image Without a VAE
AMD’s Nitro-P is a text-to-image model that works directly on pixels and drops the VAE decoder, reaching 90.02 samples/s in one-step mode on an MI300X.
AMD’s Nitro-P is a text-to-image model that works directly on pixels and drops the VAE decoder, reaching 90.02 samples/s in one-step mode on an MI300X.
AMD details a local pipeline on a single Radeon AI PRO R9700 that turns one text prompt into a playable 3D world with LLM-driven villagers.