In the foreseeable future it's not possible for AI to replicate a 2D drawing style across many images in a consistent & aesthetically pleasing manner. AI can't do it because a given 2D style is not "mathematically" consistent across images. There are slight variations with each new drawing that basically define the style. For example you can draw 10 images of essentially the same thing (e.g. a given character getting up from bed as seen from the back) and yet you can vary theoretically-unchanging details (e.g. body part proportions) and perspective WILDLY across these images without a human interpretation finding an aesthetical fault. In addition, when AI generates 2D images, the stochastic nature of the process cannot ensure sufficient consistency across multiple images so they won't look like they were drawn by the same hand. You'll basically end up with 1000 different shots of the same character in different positions but they look like an AI hallucination, not a comics strip drawn by a human.
A possible solution I would've loved exploring if I had another lifetime to spare is to use AI to generate reverse-scannable 3D models (kinda like those in pic below). You basically put in the work to generate via AI just a handful of high quality reasonably consistent images for every model. Then you put these images through a reverse-scanner that infers 3D models from 2D pictures. Then you fine-adjust the 3D models to get them in a final form and finally you start using them via classic 3D modelling/animation/rendering to create a game that looks perfectly consistent across shots. You even have the choice of going with a full 3D look (which would be very exciting in itself) or use techniques that make 3D renders look more like 2D images. You lose plenty of the 2D hand drawn feel but the gains are monumental.
