I’ll start by clarifying that the goal of the NVIDIA researchers involved in this project was not specifically stated as such, i.e. no one came up with the idea of ​​creating a generative AI model specifically designed for creating animated GIFs, which aim to become memes on social networks. However, and as you will see in this news, the truth is that this is what came out of his artificial intelligence laboratory in Toronto, Canada, to the particular happiness of all those who already consider a thousand and one to be memes. they want to create.

This is well known NVIDIA has already focused on artificial intelligence a few years agoand that they did so both from a hardware point of view, where they designed and produced chips specially equipped for the requirements of this discipline, and from a software point of view, in an area where they managed to become a benchmark with the most diverse developments, from DLSS technology to applications like GauGAN2 to many of other successful experiments and proofs of concept that extend the reach of artificial intelligence.

If you’re already familiar with GauGAN2, the technology we talked about earlier, you’ll have a better understanding of how high-resolution video synthesis with latent diffusion models works, which translates to high-resolution video synthesis with latent diffusion models. It is a model that, like GauGAN2, works with text inputs, but which uses the Stable Diffusion technology base generate short videos in high definition. More precisely, this model can generate videos of up to 4.7 seconds with a resolution that can reach up to 1280 x 2048 pixels. If the resolution is reduced, longer videos can be obtained.

To achieve this, this model combines the capacity of Stable Diffusion with the nature of latent diffusion models, which, as you can see in this case, allow a particular image to be given temporal dimensionality, the mode in which the animation is obtained with the challenge similar to the one we would use in AI to generate images, but in which we will also need to include details related to the evolution of its action.

You can check the result in the many examples that NVIDIA has published on the model information website, where you can also find a lot of technical information about how it works. In these examples you will see that at this point we are not talking about videos that could be considered completely real (although the driving simulations are amazing), but we have to remember that we are talking about incipient development, so it isYour ability to evolve is very high. And in the meantime, it’s worth starting to think about exactly what I said at the beginning, that is, about the immense number of possibilities that such a model offers in the creation of animated GIFs that can, already as memes, end up turning to the world.