I think Make My Statue is a useful tool that, for the most part, produces great results. However, I’ve noticed that it tends to be biased toward more masculine facial features, except in cases where the female subject has a very stereotypically feminine, almost Barbie-like appearance.
With other female subjects, especially those with somewhat androgynous features, the resulting statue can end up looking predominantly male, even when the photograph clearly and unambiguously depicts a woman.
My suggestion, if technically feasible, would therefore be to add a simple Male/Female checkbox or selector alongside the photograph. When Female is selected, the model could be slightly biased toward classically feminine facial characteristics, and vice versa when Male is selected — while, of course, still staying as faithful as possible to the original photograph.
The problem is difficult to solve because it is related to the availability of source data to train the AI.
Whenever you search for photos or 3d models, you will naturally find many more images that conform to certain opinions of beauty.
I do not think it is technologically infeasbile, but I do think that it’s a lot of work to build something that will not be commonly used.
That being said, it should exist and I would agree with you that I would like to see someone do the work to build this.
A way around it might be to build an AI that can rig a figure (add bones for animation). Because that will reqiure the AI to understand the location of the joints and body parts, and then those parts could be individually modified.
I agree that the availability and quality of training data is probably an important part of the problem. My suggestion was not really meant to prescribe the technical solution, though.
What I had in mind was simply that selecting Female or Male could make the system use a different prior when reconstructing the face. For example, Female could use a model, fine-tune, or training set that is primarily based on female faces or statues, while still keeping the actual photograph as the main reference.
So the intention would not be to force the face into a stereotypically female shape, but rather to let the model fall back on female facial geometry in areas where the photograph leaves some uncertainty. I would imagine that this could help especially with more androgynous faces, where the current model seems more likely to resolve that uncertainty in a masculine direction.
