Only light skinned people on the video examples. Ethnic diversity and accuracy used to be a problem with the models of the past. I wonder how the model would excel at prompts grokking at that.