Krea 2's prompt expander addresses a mismatch between detailed image-training captions and the brief requests people actually write. It turns a short request into a richer visual description before image generation.
The team first trained the expander on paired short and detailed captions, then refined it using feedback on the resulting images and checks for faithfulness to the request. Training also rewards visual variation: Krea reports that weakening this reward caused outputs to become less diverse. The aim is to make concise prompts useful while preserving room for different visual interpretations.
Terms in this piece · Glossary
transformer — The neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.
attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
