It makes things complicated, but as I experiment with the new update to the Ideogram 3.0 model, it's clear that models 2.0, 3.0 (March version) and 3.0 (May version) all have different strengths and weaknesses. Sometimes the best approach is to generate a scene in one of them and then remix it in one of the others.
The 3.0 May version seems to be trying to correct the trying-for-too-much-detail problem of the 3.0 March version (which caused the weirdly wrinkled feet problem). But so far, its biggest weakness seems to be a tendency to go too far in the other direction and make people look like plastic dolls. It will probably get refined as it "learns" through use.