Text-to-Image

An update of Google’s flagship multimodal model keeps track of colossal inputs, while an earlier version generated some questionable outputs.

The longer text-to-image models train, the better their output — but the training is costly. Researchers built a system that produced superior images after far less training.