Entering a sentence and seeing photorealistic images appear on the screen as they illustrate it, this is the feat that an Artificial Intelligence developed by NVIDIA achieves. Baptized, this is an evolved version of this AI, since had already been presented in 2019. At the time, AI already knew how to transform sketches into photorealistic images. Nvidia also offers a baptized to use this process. The only constraint is that the software requires a powerful graphics card from the brand to be able to function.
To turn words into pictures, GauGAN 2 uses a special generative AI model that has been trained from 10 million landscape images. So he doesn’t really know how to improvise if the keywords have more than one meaning or if they are too subtle, but this is only the beginning.
Version 2.0 of “Draw me a sheep” with these simple words “Paint me a picture”. NVIDIA research shows the GauGAN AI art demo now responds to words. © NVIDIA Developer
Mix keywords and sketches
Better yet, it is possible to mix the functions of Canvas and those of the keywords to combine text and. The process makes it possible to go further in the personalization of the rendering based on the sketch.
On his blog, the founder has published a short video that shows GauGAN 2 at work. On a web page, the user gradually enters keywords, such as ocean waves hitting rocks on the beach (” theof the ocean hit the rocks on the beach ”) and we see the corresponding images displayed as and when typing. Impressive ! Nvidia offers to test the AI from a but during our attempts, the did not answer.