It was possible to paint and create digital images with our words by using many different artificial intelligence applications such as Midjourney, GauGAN2, Craiyon, especially DALL-E, which was opened to everyone’s access recently. A new artificial intelligence to serve in this field was announced by Google today. The name of this artificial intelligence is Dreamfusion.
Late last year, Google announced Dream Fields, an artificial intelligence capable of rendering various objects in three dimensions. Dreamfusion, on the other hand, emerged with the combination of this artificial intelligence and Imagen, the image model artificial intelligence of Google.
Producing three-dimensional objects from two-dimensional images
“We are pleased to announce DreamFusion, our new method for Text-to-3D!
We optimized a NeRF from scratch using a pre-trained text-to-image propagation model. No need for 3D data!”
Thanks to the technology called Neural Radiance Field (NeRF), Google’s Imagen artificial intelligence creates a three-dimensional model by combining two-dimensional images in its hand. Thus, it is not necessary to train artificial intelligence on three-dimensional models. In addition, the produced models can be used together.
At least for now, the images Google offers are not as realistic as some DALL-E images. In fact, the technology giant’s product highlights are its ability to create “re-illuminated models with Lambertian lighting models that have consistent, high quality normals, surface geometry and depth”. To put it more simply, when you create an image, it is possible to create realistic and consistent images in terms of width-length, front-back, depth.
No word yet on when Dreamfusion will be available, but alternative ideas are already emerging on how it might be used in the future. It seems that this artificial intelligence can be used in many different areas, from images taken in the medical field to fully modeling the structure of the organ to be operated on, to creating models for indie game developers.
26340
