[MASSIVE RELEASE] Supra2-IMG - a tiny 100M text-to-image model - SOTA quality and open release!

Explain Like I'm Normal

SupraLabs released Supra2-IMG, a tiny 100M parameter Diffusion Transformer model trained from scratch in less than half a day. Despite its size, it produces high-quality 256x256 images and is efficient enough to run on a standard CPU in 20 seconds. This release challenges the assumption that quality text-to-image generation requires massive compute budgets and billions of parameters.