2 weeks Ago
Generative artificial intelligence models are nothing new for many of us, but the University of Surrey may have something very unique on its hands. The educator has released a new generative AI that works locally on-device. That traditionally leads to longer wait times since a model can't lean on cloud infrastructure for processing, but through some wizardry, the University of Surrey's NitroDiffusion model can generate images in real-time as a user types a prompt.
As quickly as you can type "a photograph of a meerkat floating in space, wearing sunglasses", the image is generated to those specifications. This image generator is a game-changer The code is available on GitHub , where the team explained: "We introduce NitroFusion, a fundamentally different approach to single-step diffusion that achieves high-quality generation through a dynamic adversarial framework." "While one-step methods offer dramatic speed advantages, they typically suffer from quality degradation compared to their multi-step counterparts.
" The approach they have taken leans on the concept of art critics providing comprehensive feedback on composition, color and technique. It maintains a large pool of specialized discriminator heads that guide the generation process. In essence, then, consider the model to be packed full of mini, AI art critics that help guide the generation as words are entered via the prompt.
According to the team, these discriminators also lend their increased knowledge back into the discriminator pool once they're sampled, meaning the model should improve over time. More from Tom's Guide.
