Rock band Abrahadabra wanted to do something special for their debut EP. They wanted to create visual artwork that would go well with their overall artistic vision, but at the same time, they wanted to make something unique and something that will really stand out. So, here is what we did to achieve that, making them Abrahadabra one of the first bands in the world that used AI for the majority of their visual and sound representation.
Are you afraid that AI might take your job? Make sure you are the one who is building it.
STAY RELEVANT IN THE RISING AI INDUSTRY! 🖖
We started with the image. The band is composed of two members and they are actually a couple. So they picked up one of the images of members taken back in 2012. by Dejana Batalovic. It seems that this image is their friends’ favorite and it even ended up on their friend’s fridge. They really love this photo, so we decided to use it as our starting point.
We wanted to apply a technique called neural style transfer to this image. In a nutshell, this is a technique where deep neural networks are used to learn a style of a painter and then apply that style to the content image. Style transfer is a fun and interesting technique that showcases the capabilities of neural networks. More about this method can be found in this paper.
The mother of one of the band members was an amateur painter, so the band wanted to apply her style to this image. This sounded really cool because the band explored themes of connection, heritage, and resurrection in their music. This fitted their artistic vision perfectly. Here are some of Vera Živković’s paintings that we used:
So, we run a couple of tests, learning styles from different paintings and applying them to the image. Here are some of the results:
The band decided that they will use a blue version of the image, but other versions were not done in vain. These versions were used for creating lyrics videos, but more on that later. Finally, our designers cropped the image and created a band logo for it. The final version of the cover looks like this:
‘Dole u Srbiji’ Video
The band wanted to do something extraordinary for their lead single Dole u Srbiji, which is a commentary on polarized Serbian society. Again, they wanted to emphasize themes of heritage, empathy, and love/hate relationship with their homeland. We came up with the idea of using Deep Fakes with influential and famous Serbian scientists, painters, poets, and so on, singing the lyrics of the song. We used a technique called First Order Model, on which you can find out more in this paper. This method really blew us away last year and we were really anxious to do something meaningful with it.
So we used a list of more than 50 famous Serbians including Nikola Tesla, Mihailo Pupin, Vuk Karadžić, etc. We got their portraits and then run a lot of tests:
Some of the tests failed in really funny ways:
Finally, we picked up the best results and combined them in one video. Put some VHS 80’s effects on top of them and voila:
Finally, we decided to add some bonus to this and utilize styled images that we created for the cover. Although this isn’t really an AI technique, it turned out to be a really interesting approach. We used affine transformation to move between different versions of the EP cover to create this moving and “breathing” effect.
We stacked them together to create a longer video. This is used for lyrics videos, again combined with the 80’s VHS effects and lyrics in a form of a subtitle. This created a really interesting future-retro look. After all, the roots of AI can be traced to the ’80s.
This is an interesting showcase in which we utilized a lot of techniques that we don’t get to use so often. The combination of art and AI is really interesting especially combined with the 80’s look and feel. This gave the whole project a retro-future vibe and it was a pleasure to create something like this.
You can watch all the videos here.
Abrahadabra – “Više nas spaja nego razdvaja” EP is available on all popular platforms.
More info can be found on Abrahadabra official website.
Thank you for reading!
Building Smart Apps
Read our blog posts here.