My 2nd creative project for DIG 5831, Computational Media. A creative visual work using generative ai techniques.
This project is exploring how generative AI can understand and visually represent human emotions. I used text to image AI tools to generate the images found in this project. The pictures produced can be viewed in the video in the README below, and at the project website. Website can be found at: https://danyabaron.github.io/visual-gen/
I primarily used DreamStudio (https://beta.dreamstudio.ai/), and Adobe Firefly (https://firefly.adobe.com/generate/images) for this project. The settings for each image can be found under the 'configuration settings' folder. Each image file has been named appropriately so the name will also match the settings image files.
Comp.Media.Creative.project.mp4
In this project, my goal was to explore how AI perceives human emotions, and how the AI relays that through image generation. I focused on emotions that may have been a bit more difficult than just "happy" or "sad", and explored more complicated emotions such as determination, frustration, inspiration, surprise, and more. I wanted to create art that portrayed our emotions in a way that AI viewed. Some of images that I picked may have an underexaggeration of the emotion, or overexaggeration of the emotion. I did this on purpose to show how AI perceives us and our emotions. I have always found art to represent the human experience, almost acting like a mirror that reflects towards us. Growing up, I often frequented art museums with my family, so art has been a very integral part of my life. I have to be honest, I don't love AI generated art, simply because I feel it lacks the depth of representing human emotions. I find that AI art is really good for digital imaging for maybe websites, video games, and other animated images that could be used as a tool for someone's digital projects. But for representing the human condition, experience, and emotions, I personally feel that AI cannot produce the same quality of art that humans can. Art is a way for us to see our emotions displayed either in movies, text, or images. I just feel that since AI is a machine, it cannot properly articulate the very complex emotions and feelings that we as humans have. That's why I explored this for my creative project, to test my hypothesis on how AI views our emotions, and how it creates art about our emotions. I also looked at some artists' artwork in the AI sphere, and became inspired by Mario Klingemann's "Appropriate Response", which was done using a custom made GPT-2 model.
Rather than giving prompts to the AI such as, "show me a happy person" or "show me a sad image", I tried to create specific prompts to properly obtain the emotion I was trying to depict using the generative AI. I think the best prompt that the AI understood the most was this one: "a lone figure sitting on park bench with fog around them". I wanted to create this melancholic image that represented the emotion loneliness. The AI gave me a few options that all were very good, but I chose the image of the person slumped into the bench, as I feel like this image truly represented what I was going for. I was surprised that the AI did the prompt so well, and portrayed the emotion I wanted. I also gave the prompt of "a person frustrated with themself", which I thought at first was not very specific, but then the AI generated an image that I felt perfectly portrays what that prompt feels like. I was surprised by this.
I found that the AI seemed to be better at understanding emotions that are usually viewed negatively in society, primarily loneliness, frustration, and anger. All the prompts that I did for this image came out very well, and depicted the emotion clearly.
I found that I had a more difficult time getting the AI to represent more positive/neutral emotions such as joy, determination, surprise, love, and inspiration. The surprise prompt that I picked was: "a person winning lottery tickets, eyes wide with disbelief and surprise". At first, this was just disbelief, but then I added the "surprise" keyword because the images that were producing were not accurate at all. I believe the AI also got confused with the lottery tickets as well. I also tried to depict determination as an emotion. The prompt I chose for determination was: "A lone runner pushing themselves to the finish line of a marathon, sunlight glinting off sweat". I experimented with other prompts as well but I found that this one had the "best" image of what I was going for, even though from first glance it is difficult to see that the emotion I was trying to do was determination.
All in all, this project explores how the AI interpreted my prompts that were specifically designed to depict emotions such as frustration, arguments, joy, love, inspiration, determination, surprise, peace, and disappointment.