Generate images showing each prompt token’s visual impact
demonstrating how the model retrieve words by sub-word token