Jasper AI's MONET dataset offers 104.9M curated images for AI

Jasper AI has unveiled MONET, a massive open-source dataset comprising 104.9 million high-quality, curated images with corresponding captions and metadata. The dataset was meticulously filtered from an initial pool of 2.9 billion images, ensuring only the highest quality samples remain. Licensed under Apache 2.0, MONET is available on Hugging Face, making it accessible to the entire AI research community.

MONET's creation process is detailed in an accompanying paper, offering transparency and reproducibility. Beyond the dataset itself, Jasper AI provides three companion projects: a UMAP visualization for exploring distribution, a retrieval tool for text or image search, and a codebase for training text-to-image models directly on MONET. These tools lower the barrier for both research and production use.

For practitioners, MONET addresses a critical need for large-scale, clean, and well-documented image-text data. Unlike many datasets that require significant cleaning, MONET comes ready to use, saving teams weeks of preprocessing. Its size and quality make it suitable for training state-of-the-art generative models, improving multimodal understanding, or fine-tuning existing architectures.