These high-definition faces, and yes, even cats, can be created as 3D pictures using a new AI
Artificial intelligence and neural network-produced realistic graphics are now all the rage. Pictures with scary but oddly realistic takes, such as these AI-generated Duke Nukems, have gone viral.
While AIs like DALLE-2 can do Photoshop tricks better than humans, things are going to become a lot crazier when they transition to 3D.
A common machine-learning technique called GAN puts two neural networks against one another to produce predictions that are more precise.
It has previously been proven successful for reproducing 2D photos, and according to MarketTechPost, Stanford researchers have used it to produce potent 3D images.
These experts refer to this AI model as EG3D.
It transforms 2D GAN photos with high quality into 3D images. It boasts of being able to handle it on laptops and does this in a manner that is both far less computer-intensive and more accurate than prior efforts.
Github does suggest 1–8 high-end Nvidia GPUs, but you could probably get away with a decent laptop for the task.
As a result, the team has created some unsettlingly realistic images of people's faces taken from various perspectives.
Additionally, it has produced some adorably adorable cat images, which makes it seem much less dangerous.
The major drawback of these pieces is that they might be challenging to modify and improve. They aren't quite Blender models that are ready to be altered according to the user's desires since they are generated by AI.
The next stage of these technologies will probably benefit greatly from a separate machine-learning model called GiraffeHD, created by the University of Wisconsin and enables various factors in these photos to be identified and picked.
However, given how these 3D graphics are produced, it could take some time before we see them in video games.
Although it's still a ways off, the ability to create 3D photos from previously created 2D images might be a significant windfall for the business.
However, the chance of seeing these AI-generated images in the media will increase significantly if it is possible to make these models function in applications that regular users can use.