Back to Articles
Ressources The Problem: A Data Gap Holding Back Text to Image research The Curation Pipeline: From 2.9 Billion URLs to 104.9 Million High-Quality Images Content Distribution Captioning: creating some text for every images. Adding synthetic data Validation Limitations Training Your Own Model with nano-t2i Conclusion Jasper research is releasing MONET, the largest open, image–text dataset ever released. It was built from 2.9 billion images and refined to 104.9 million high-quality samples.
The launch comes with nano-t2i a minimal codebase to train a competitive diffusion model from scratch on a single GPU in a couple of days.
Together, these give researchers everything they need to train production-grade text-to-image models without the prohibitive cost and complexity that has long gatekept the field.
Ressources
