i1: A Simple and Fully Open Recipe for Strong Text-to-Image Models
Researchers have developed "i1," a 3-billion parameter text-to-image diffusion model that matches leading performance while remaining fully open-source. Through extensive experimentation, the team identified key design choices and training insights, such as equal weighting for dataset mixing and the benefit of larger text encoder adapters. The i1 model, trained on publicly available datasets, significantly outperforms existing open-source models and provides checkpoints, code, and a data processing pipeline to foster future open research. AI
IMPACT Establishes a practical foundation for open research in text-to-image diffusion models, potentially accelerating community development.