We propose a novel modular framework that learns to dynamically mix low-rank adapters (LoRAs) to improve visual analogy learning, enabling flexible and generalizable image edits based on example transformations. - View it on GitHub