Biggest SCAM in the AI ART or new possibilities for everyone?
Hi everyone. A few days ago I was testing one of the neural networks for AI ART in order to find options of generating some architectural facade elements for a video mapping show. After several attempts, I generated both individual facades and single elements, sets of architectural elements.
What AI gave me over time shocked me a little.
For example, upon request Set_of_columns_architectural_facade_elements_in_Art_Nouveau_style__Front_view__Isolated
I saw some generated pictures showing the graphic canvas, previews of pictures from some photo stocks. Also there were watermarks on some of them. But the top class was shown when AI began to generate images with a specific logo of the site, from where, most likely, the content was “pulled”, for its use in the dataset for training the neural network. I have no other guess. Here, see for yourself.
Examples with watermarks…
Here you could see an explicit Alamy photo stock logo. On the net I saw some similar pictures that were created by leather bags, protein carriers… ))))
It is clear that to train a neural network, you need large data sets, which are later formed into a dataset. Writing a few lines of code, or using a parser to capture all the preview images on a large photo stock site. I think it is not a big deal for developers. And most likely, resolutions of small pictures will be enough to train the neural network…
In October 2022, some photo stocks have already begun to sound the alarm, notifying their authors that they will no longer accept works that were generated by AI art systems, because, as they write, the origin of the works is not clear and there are unresolved copyright issues.
Photo stocks protect copyright holders and their clients, respect copyrights, etc. But this is partly, and only at first glance.
I think it will be beneficial for them only until they themselves master this technology and create some kind of their own commercial service based on their library of materials.
After all, some photo stocks already force you to upload a release model, which directly says that the model entitles the photographer and the photo stock to its biometrics, and on the basis of this biometrics it is allowed to generate other faces, people, pictures, etc. in the future. Think about it…
And here the issue gets even more complicated. Indeed, on photo stocks, if you want to sell photos of locations, you also need a property release signed by owners of this location, who gave you the right to shoot. But if such a photo is sold as an Editorial, then such a release is not really needed. And the picture will be pulled into an AI dataset without the consent of at least 2 copyright holders. And if you think even deeper, then at least 4.
- The architect who came up with the plan and design of the building.
- Designers and sculptors who created elements for the facade.
- The owner of this location.
- The photographer who took the picture.
Some examples are very beautiful and with more detailed iterations it is possible to achieve the desired results. In any case, I believe that AI is the future. And it needs to be adopted by absolutely everyone. But on other hands – respect intellectual property and copyrights. Find Solutions how to manage it and you become very rich.
Alexander Kuiava, a new media artist, content producer, CEO of Austrian video production studio LIME ART GROUP
Author of self techniques for developing video projection shows and a teacher at the Frontskill School of New Media Art. Online entrepreneur and social investor. VJ in retirement.