The Dereverb model was trained on dry sounds and the same sounds with added reverb (wet sound). The model then learned to generate the dry sound based on the wet sound. This is the same idea as stem splitting.Start with Remove Reverb, or use the Dereverb preset from Studio.