Sorry, not sure how to ! post so it opens in your instance.
TL;DR
Any result is going to be biased. If it generated a crab wearing liederhosen, it’s obviously a bias towards crabs. You can’t not have a biased output because the prompting is controlling the bias. There’s no cause for concern here. The model is outputting by default the general trend of the data it was trained with. If it was trained with crabs, it would be generating crab-like images.