Some works generated with artificial intelligence (AI) applications such as DALL E, Stable Diffusion and Midjourney have been winners of creative contests. The problem is that in certain cases the jury thought which was rewarding the work of people and not from a set of well-trained algorithms.
We are witnessing how AI presents opportunities and challenges in a wide variety of fields, one of them being the artistic field. In this sense, researchers from the University of Chicago have presented an interesting application that claims to protect the creative works made by people.
Glaze, a tool that could please many artists
The app developed by computer science teacher Ben Zhao and his students has been baptized as Glaze. However, to know how it works, it is necessary to remember how AI imagers are trained. These systems, although they receive a promptwith prompts, do not create works “out of nothing”.
As we mentioned in the middle of last year, they work with huge datasets that contain copyrighted images. This training is the basis for the generative models to work, that is, they have a starting point to do their work. Now this also worries many artists.
The epicenter of the issue is not only the fact that their works are used to generate new works in apps that are often paid for, but they imitate his style. Some artists, according to ARTnewsthey were surprised to see LensaAI create images with what could be their signatures.
Original creation by artist Karla Ortiz (left), creation created by an AI model (center), and creation made by an AI after cloaking the original art (right)
According to LensaAI, the app’s generative AI imitates paintings and invents signatures, but they are not actual signatures of people. Whatever the situation, there are artists who want to prevent their works from fitting by training AI models. For them, the application that we mentioned above comes into play, Glaze.
So how does it work? Let’s imagine that an artist wants to protect her creations published online. Nothing would prevent that through techniques of web scraping if I work ended up being part of a data set. However, an alternative may be that this artist passes her works through the tool created by the researchers before publishing them.
How Glaze works
Glaze, they explain, adds small changes to the original artwork. These are modifications that are not visible to the human eye, but prevent AI models from being able to reproduce their creation. Let’s look at an example supplied by the researchers. At the top of the image we see a creation that has not been protected by Glaze.
The Stable Diffusion generative model chosen for this test successfully learns the art style. At the bottom, however, we see how the image has been subjected to Glaze and the model cannot meet its objective. It collects data, yes, but in a completely different “style” than the original artist.
For now, we have to wait to use Glaze. Its creators say they will release versions for Mac and Windows in “the next few weeks.” They also promise that the tool will be totally free. If you are one of those interested in trying it, you can see the news at the page dedicated to the University of Chicago project.
Images: University of Chicago
In Xataka: We have tested the new Bing with the improved version of ChatGPT: Google has reason to panic
