Seminar: Cartes Blanches (2021–23)
How to Visualize Dreams
Jana Wüst
In my project I want to visualize my own dreams. From an early age on I write a diary, which often includes memories of my dreams too, and I used to always wonder, if it will ever be possible to visualize memories and dreams. This is where deeplearning tools come in. In my project I use three different tools: text2image, a neural network that utilizes CLIP to steer BigGAN towards images that match a textual prompt in any language. I also use ArtBreeder, a machine learning-based art website, using the models StyleGAN and BigGAN, which allows users to generate and modify images of faces, landscapes, animals, paintings and other objects. The third one is Spade Coco, a neutral network that uses sketches in certain color-codes and translates them into realistic images.
The images generated with these deep learning tools are often unclear, blurry and quite abstract. Dreams, too, are often blurry in memory, not quite logical, not quite chronological in their structure, very abstract, surreal, and unrealistic. Because of their visual characteristics, it is why the above-mentioned tools are ideal for the implementation of my project.
I chose to transpose four different dreams that I had in the year 2021 so far, because the average person has four to six dreams per night, even if memories of them are missing after waking up. I exclusively use generated images to visualize the dreams. In addition to that I generated a voice, using a free text2speech online tool, that narrates the story. Lastly, I added sounds and soundscapes to bring the dreams more to life. The animations have a collage-like character, which represents the collage-like memories we often have of our dreams.
Nevertheless, it is not the dreams that should be in the foreground, nor the analysis of them, but the examination and exploration of the possibilities and limits of the four deep learning tools text2image, ArtBreeder, Spade Coco and text2speech. Detached from my personal experiences and thought of in a next step, this implementation can be a hypothetical future tool, with which each person can visualize their own dreams.