• 

  Final Artwork

Code notebook: http://tiny.cc/19notebook Presentation (short): http://tiny.cc/19short Presentation (detailed): http://tiny.cc/19Detailed

/

WHAT

We used random text generator to create a prompt narrative for the art piece; We wanted the machine to take the lead in taking all the storytelling decision from verbal to visual.
We analysed and dramatized the main elements in that narrative, and started working from there.

WHAT

The generated narrative had two main characteristics, it was symbolic and mystical. That gave us a clear aesthetic direction to make it painting like and surreal and exaggerated. We wanted to show the misery of that lady, crying in the desert. With nothing to eat and no place to go. We emphasized that with the stone inedible fruits. While she was crying tears of blood, a mercy came taking the shape of a horse like creature to rescue her.

HOW

1.Dataset collection: we scraped the horses dataset from the internet, resized the images and used data augmentation to increase the diversity of our data.

2.Choosing the model.We wanted to used StyleGAN2 to generate the narrative elements to emphasize that it’s machine imagined narrative, but for time limitation we ended up creating the main figure.

1. Creating the 3D mesh. We initially tried a Pixel2Mesh++ but the code was deprecated. So we tried to find a computational solution. So we illuminated the background and convert it to grayscale then inverted the result. And then we used PIL to extract the pixel data, RGB and alpha. Then we planned to use Matplotlib to create the 3D mesh, but we failed in converting the pixel data to xyz arrays.

2. UV mapping. To further include the machine in the process of creating the horse character, we used a self-organizing map algorithm to generate random blocks of color which we processed in photoshop and finally mapped onto the 3D mesh.

Built With

• glob
• numpy
• pil
• python
• selenium
• stylegan