Another option is to have a 2-4 skin tones and just randomly select one for each respondent. It's introducing another variable but presumably you could collapse over it.