The design process detailed in full below, from initial GAN work to refined designs.
Please click on an image to enlarge.
To start the process of creating shapes from a generative adversarial network (GAN) I needed to create a dataset. Typically these will need to be over 1000 images, so I decided to experiment with a smaller dataset. Starting with around 150 images.
Initial GAN-generated imagery. 150 image dataset.
I then decided with my collaborator that we should create more data, taking us to around 350 images. The gif below shows the data once we had gone over the 300 image mark.
Using these shapes, I started to create silhouettes.
Video showing the versatility of using AI-aided design.
Refining my designs.
I then went back and looked at collaging the silhouette imagery to elevate my designs.
I chose some full designs to collage.
Matching collaged silhouettes to collaged designs.
Choosing my favourite imagery.
From this analysis of my designs, I went back and drew a 10 person line-up.
At this point, I had also started to learn CLO3D and had chosen this design to 3D render.
