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

In three linesMicrosoft Research presents Lens, a text-to-image model with 3.8 billion parameters that matches much larger rivals on benchmarks at a fraction of training cost. Key innovation: 800 million detailed captions generated by GPT-4.1 instead of vague web alt-text. Code and weights released under open-source license.

Summary generated by Claude — human-verified