Throughout history, the aesthetic of body image and the definition of the “ideal body” have always changed depending on different eras and social backgrounds. Beauty standards from ancient Egypt to the present day have been shaped by cultures from both the East and the West. People consistently describe the “ideal” body image as iconic based on different cultural norms, perceptions, opinions, standards, and preferences.

Nowadays, the world revolves around data, numbers, information, and forms. Using techniques to make data visible is what data visualization does. The information can contain various types of content. It is not limited to numbers. Image dataset means images can also be a dataset to collect, categorize and analyze. Generative Adversarial Networks (GANs) are a type of generative modeling that utilizes deep learning techniques, including convolutional neural networks (CNNs), as well as an adversarial training process to generate new data that resembles the original training data.

Based on my hypothesis about artificial intelligence beauty bias, the design approach of my experiment is to test this beauty bias and find method to improve this experience, or in other word to say, find a way to find the fairness, diversity and inclusive. In contrast to my original plan of manually analyzing body images for beauty, I incorporate existing machine-learning models to assist with the critique process. This approach still aims at identifying and addressing beauty bias. However, the questions have become more focused on specific aspects such as skin tone, body shape, or clothing texture.

The aim of this data visualization is to provide a guideline about the prompt word used on Midjourny.

The primary prompt was “Fashion model beauty” and “Posthumanism and Futuristic” as an art style. Keep inputting the command to add and remove synonyms and related key terms to see what could emerge after changing. Also, set the background as flowers to distinguish differences and the content styles.

I tested the listed keywords from Circular dendrogram, as these are cascade related. However, only some of the keywords produced the desired results. The images were reorganized and rearranged so that viewers can observe and compare the differences between these artworks. Images and prompts are produced after filter layers. In the picture, (Figure. 34)  shown here, see some initial input words and phrases such as “fashion magazine mode” and “Posthumanism fashion model beauty,” “Posthumanism Beauty.” By adding related words like diverse, and aging, we get other aspects of beauty. So, discover the images’ color, skin tone, and body shape as the primary focus. 

The Midjourney itself exerts stereotypes, and it somehow displays specific “disrespectful” photos without the ability to identify them. But all of this comes from learning from its original initial dataset of thousands of photos. In other words, AI bias is also human bias. The process and result of AI learning from an extensive database of images to produce new art paintings is an innovative form of art creation. The artists can filter image details through input and output. Therefore, humans remain the most crucial factor and the main source of decision-making. With prompt keywords to input, we need to be specific on what, when, and how. The keywords should be as specific as possible to avoid Image bias and stereotypes. Because the AI do not have imagination on create image.