Microsoft Research's Lens proves detailed captions matter more than raw scale for training efficient image generators

Microsoft has introduced Lens, a text-to-image model featuring 3.8 billion parameters that competes effectively with larger models while significantly reducing training costs. The model's success stems from utilizing 800 million detailed image captions generated by GPT-4.1, which provide richer context than typical web alt-text. By making the code and weights openly available under an open-source license, Microsoft encourages developers and researchers to leverage this innovative approach in their own projects.