OK, now with Autodesk’s demise being imminent, from what I’m reading on these forums 
, I was wondering whether there’s any effort to put in features like Autodesk’s Dreamcatcher has, into Blender? :
https://autodeskresearch.com/projects/dreamcatcher
ie. the software actually makes the model, or part of the model, FOR you??!!!
By the time AD perfects it , they wont need to sell licenses anymore. They can just generate one billion artists working for free 24/7. So anyone who thinks this is great is going to have a short ride.
Yes, some sort of machine learning assisted workflow is the path for content creation going forward. For real world creatures and objects, photogrammetry combined with machine learning has huge potential.
For non-real-world stuff, I’m convinced we will end up having artists just draw in a 3d space (as in line art) and have machine learning be trained to generate/modify a mesh on the-fly based upon its interpretation of said line art. Thus turning all concept artists into 3d modellers/sculptors, and of course speed up the testing/refining stage immensely.
As for Blender, there’s a lot of low hanging fruit to tackle at the moment, but with the funding at their disposal nowadays, I definitely think this is an area they should look into as soon as they can. At least hiring someone with ML experience to start making plans and experiments to see how it can be incorporated into Blender.
Maybe time will prove me wrong, but I can only see this useful for idea stage - playing around with possible forms. Something for production is “never” about one part, but how many parts interact and fit with eachother. In the rare cases of one part, it’s usually about cutting the production cost maximally and shapes being the result of design requirements (excl enclosures which may have more aesthetics to them).
The link you provided makes it sound like it something focused on product design. Giving the software a series of constraints for the product (ex: type of material, amount of weight it needs to hold, etc…) and getting back some sort of “physical” design that could be adjusted and later built into a real world product. It doesn’t sound like software that is meant to do anything creative like character modeling or whatever.
Edit: I could see something similar working in blender, but it wouldn’t anything AI based. It would be something like kitbashing where every part has a large series of adjustable parameters for appearance (parametric objects).
I think they already have.
Perhaps you need this