📰 [Feb 2025]

We release TextAtlas5M, a dataset specifically designed for training and evaluating multimodal generation models on dense-text image generation.