Abstract
Generative artificial intelligence is an emerging technology that has impacted society recently and with productivity gains in the coming years. This technology learns from existing data to generate high-quality content such as text, images, music, speech, videos, and code. As part of the creative industries, collaboration between generative algorithms and human creativity brings challenges due to producing content and artistic images in response to natural language requests. Thus, this chapter establishes a methodology based on text-to-image generative artificial intelligence for producing assisted art in contextualized cultural content to generate unique and diverse images of the historical-cultural identity of southern Brazil. The steps of the methodology that can portray elements of Brazil's biome, cultural, and historical events are dataset assembling, model training and fine-tuning, and content enhancement and post-processing. In addition, the methodology allows an immersed experience with people being transported in period clothing and within a historical or cultural setting.
TopIntroduction
Generative Artificial Intelligence (GenAI) is one of the top 10 emerging technologies reported by the World Economic Forum that impact society in the next three to five years (World Economic Forum, 2023). GenAI is a type of deep-learning model that doesn’t require knowledge or entering code but can learn from existing data to generate a variety of high-quality realistic content such as text, images, music, speech, videos, software code, or others. A GenAI model differs from a predictive machine learning model, which performs a discrimination behavior, solving classification or regression problems. Thus, GenAI models use techniques that produce or create information on the transformed data rather than simply analyzing or acting on the data they trained on.
GenAI has the potential to disrupt several industries. Healthcare applies as a tool for accelerating drug discovery, supporting clinicians to transcribe patient consultations, generating drug molecules (Anstine & Isayev, 2023), and analyzing brain activity data to produce drawings of the objects that human participants are holding in mind (Takagi & Nishimoto, 2023). Environmental science predicts weather patterns and simulates the effects of climate change. Education integrates into teaching practices or training students (Baidoo-Anu & Owusu Ansah, 2023) to develop customized learning materials that cater to students’ learning styles. Finance analyzes market patterns, anticipates stock market trends, and assists financial analysts. In advertising, creates new advertisements based on existing ones, making it easier for companies to reach new audiences. Art and design help artists and designers create new works by generating new ideas and concepts. Entertainment creates new video games, movies, and TV shows, making it easier for content creators to reach new audiences (World Economic Forum, 2023). In the workplace, increase productivity and quality, restructuring human tasks towards idea generation and editing (Shakked & Whitney, 2023).
In 2018, the professional art world was upended when the renowned Christie’s auction house sold an AI-augmented work, “Portrait of Edmond Belamy” for the unexpected sum of $435,000. That sale, which came with the tacit imprimatur of the established art community, generated much gnashing of teeth and hand-wringing in the arts sector over what artificial intelligence means for the creative industry (Stanford University Human-Centered Artificial Intelligence, 2023). A 2022 McKinsey survey shows that AI adoption has more than doubled over the past five years, and investment in AI is increasing quickly. Generative AI tools, like ChatGPT and DALL-E, improve jobs and have the potential to achieve this (McKinsey Global Publishing, 2023).
In the context of the creative industries, a collaboration between intelligent algorithms and human creativity brings challenges. Creativeness defines the ability to produce original and unusual ideas or to make something new or imaginative. Creative tasks generally require some degree of original thinking, extensive experience, and an audience understanding. Today, GenAI associates human creativity and artistic practice relevant to the creative industry due to producing content and artistic images in response to natural language requests. Research conducted by Adobe to understand the role technology plays in their creative process revealed that three-quarters of artists in the US, UK, Germany, and Japan would consider using AI tools as assistants in areas such as image search, editing, and other non-creative tasks (Pfeiffer Consulting, 2018). Creative professionals participating in this research feel that AI is a technology transformation, but they don't know how it will change their work. Over 60% of respondents said these developments will impact quite a lot, and only 5% think they will have no impact. The number of interviewees interested in having a head-start in AI is even higher, at 73%. Respondents in Japan, in particular, showed great enthusiasm and receptivity for developments based on AI.
Key Terms in this Chapter
Latent Diffusion Model: Also known as diffusion probabilistic models or score-based generative models, are a class of latent variable generative models. Consists of three major components: the forward process, the reverse process, and the sampling procedure. The goal of Diffusion models is to learn a diffusion process that generates the probability distribution of a given dataset. They learn the latent structure of a dataset by modeling the way in which data points diffuse through their latent space.
Generative Artificial Intelligence: Artificial intelligence capable of producing novel and realistic content across a broad spectrum (texts, images, audio, video, code) for various domains based on basic user prompts.
DreamBooth: Is a training methodology used in text-to-image generation model that updates the entire Diffusion model by training on just a few images of a subject or style. It works by associating a special word in the prompt with the example images.
Fine-Tuning: Approach to transfer learning in which the weights of a pre-trained model are trained on new data. Is done on the entire neural network, or on only a subset of its layers, in which case the layers that are not being fine-tuned are frozen.
Text-to-Image: Generation systems based on deep generative models for creating digital images and artworks. Given an input prompt in natural language, these generative systems are able to synthesize digital images of high aesthetic quality.
Deep Generative Models: Neural networks with many hidden layers trained to approximate complicated, high-dimensional probability distributions. The goal is to learn an unknown or intractable probability distribution from a typically small number of independent and identically distributed samples.
Inpainting: Task of reconstructing regions in an image. It is an important problem in computer vision and an essential functionality in many imaging and graphics applications, e.g. object removal, image restoration, manipulation, re-targeting, compositing, and image-based rendering.