The Renaissance of AI Art in 2022
By now Artificial Intelligence has become a significant component of technological advancement as it is employed in a vast number of industries and services. Since the invention of computer graphics, the voyage for AI-generated art was given cumulated efforts and in this year, 2022, we have made several breakthroughs. This article narrates briefly, the historical and technical details of AI art, comprehensible to the ordinary person, and its impact on the futures of targeted domains. This article also addresses my personal views regarding the insurgence of corruptive AI practices.
A brief history of AI Art
The quest for AI art first began with a technology called Neural Style Transfer (NST). In this process, we take two images, one a photograph of preference and the second, an image of a painting of a given style. In NST we ‘apply’ the latter image’s artistic style to the former through the use of a Neural Network (A typical form of software used in Artificial Intelligence) and produce a new piece of art. The below picture is from the first paper published regarding NST.
We can see how the first image of the buildings is drawn using the artistic styles of famous art pieces such as Vincent Van Gough’s Starry Night.
Even though NST created art, it needed two input images and could not create art independently. We needed something that could produce new data after sampling some given data. Hence GANs were introduced.
Generative Adversarial Networks or GANs for short are a type of Neural Network which can generate new data from existing data. This technology soon became popular with its most famous applications such as NVIDIA’s GauGAN and AnimeGAN.
The famous website called thispersondoesnotexist.com is a platform where you get randomly generated human faces with the help of GAN technology. This is where things begin to get creepy. But still, these systems were highly dependent on visual inputs. The need for a method to generate sophisticated, utterly abstract visual graphics on textual command was more relevant ever than before.
Diffusion Models, the singularity of AI art
The hottest and latest trend in AI art is the use of Diffusion Models. In simple terms diffusion models get as input a distorted (or diffused) system and learn to reverse it back to its original state. Let’s get a bit technical and dive into the inner workings of these mystery devices.
Diffusion models create new images by starting from random noise (diffused image) and performing convolutions (a complex mathematical operation) on it to synthesize the required image. Before use, the model needs to be ‘trained’ on some sample data. We gather a sequence of training data images and pass each image through a series of units which apply a little bit of noise to each image. This sequence is called a Markov Chain.
In a Markov chain, the current unit’s output only depends on its previous unit. This mathematical construct is often used in predicting the weather. After sufficient passes through the Markov chain, the model learns to reverse this process to start from random noise and finish from a high-resolution meaningful image. Predicting the final image given random noise is done via Convolutional Neural Networks (CNNs).
Soon after the discovery of this new approach, it became popular among tech pillars and soon started commercial application.
The Medici family of AI Art
In the years 2020–2022 the research on AI-generated art was in silent progress until in 2022 we got to experience a burst of new technologies. The talk about AI art surged exponentially this year with the rise of DALL-E developed by OpenAI. In 2022 its improved version DALL-E 2 was released as beta and open to the public where you can input some text and the system will generate four images based on your prompt. Soon this platform broke the internet. Below are some images generated using DALLE-2 (source: https://openai.com/dall-e-2/ )
The beauty of this system was that it could generate unimaginable depictions with such creativity and variety. It had the imaginative capacity of a human (or more so) and performed well in rendering those ideas into a graphical output. You can register for DALL-E 2 today and generate your own images for the cost of credits.
After DALL-E, the next pillar to enter the AI art gallery was Midjourney. Using Discord to attract a variety of enthusiasts, this new platform also created high-resolution images using text prompts. Below are images generated using Midjourney AI.
Next came a significant revolution in AI history. Stability AI, a company who were advancing in similar image generation methodology released its own blend of the new technology as Stable Diffusion. Not only developing a powerful generative AI but also open-sourcing the said project, put the impactful yet quite destructive, powers of creation into the people’s hands. Open-sourcing is when the source code of the software is published openly so that anyone with a sufficient understanding can mutate it and use it for their purposes. This is where things start to take a bad turn. Stable Diffusion became so popular that people started making their own variants of it.
The Renaissance
Soon an Adobe Photoshop plugin was released powered by Stable Diffusion, to manipulate images with a single text prompt. Augmented Reality applications were developed using Stable Diffusion as a utility to render AR art in real-time. Also, Stable Diffusion could support video, making it capable of generating cartoons and animations. Social media became infested with AI-powered face filters, and several apps were started for the whole purpose of generating AI avatars.
By now you most probably have used an AI-powered filter or tried out an AI art app. Apps like Dawn.AI are crafters of AI-generated avatars. Leonardo.AI is a company that specialized in creating AI-generated video game assets. And by the time I am writing this article, there is an immensely popular AI filter circulating on TikTok.
The uses of AI-generated art have extended beyond our horizons. People who cannot draw can create digital art for themselves. Tools like DALL-E are ideal for creating stock images, thumbnails and other graphic assets. Resolution rescaling and restoring damaged images is child’s play. AI can also help in providing inspiration for graphical design. The following is an output created by Midjourney which depicts a UI/UX design.
The capabilities and applications of AI art are endless, simultaneously making it a godly creative tool and a weapon of great destruction.
From Holy Grail to Sinful Cancer
By now generative AI has caused enough catastrophes. While quite early technologies like Deep Fakes have contributed to serious defamation and impersonification issues, since this is an article about art, I will preserve specific relevancy regarding the AI-art domain.
Since Stable Diffusion was publicly open-sourced, average tech enthusiasts have toyed with this software, by feeding it original copyrighted artwork of professional artists, to recreate their style. This AI-art stealing frenzy has caused massive chaos on social media as the original artists were never credited, and their art was used without permission. This all boils down to the ultimate issue where an artist's individual authentic art style is adapted by a machine and is capable of synthesizing infinite amounts of artwork with no effort, endangering the artist’s career.
Artists Sam Yang, Greg Rutkowski and Karla Ortiz had their art-style ‘stolen’ and cyber-bullied by AI artists on social media. Soon after the death of Kim Jung Gi, his artwork was fed into an AI without the permission of his family. You can read more about it here.
With the rise of AI art in 2022, we got to experience AI art competitions, where the competitors had to generate art using a prompt. This was fair under that specific context but highly unethical when people started submitting AI art for real-life art competitions. The first such incident was reported this year as a submission to the Colorado State Fair Art show titled ‘Theatre D’Opera Spatial’. Jason M. Allen submitted a piece of art generated using Midjourney and won first place, spreading controversy that an AI won a human art competition.
Also later this year, there was an incident where ArtStation, a popular art-sharing platform featured AI art and caused a massive backlash and protest from its users. You may read the full story here.
Personal Takes
Being an enthusiast of Artificial Intelligence and a hobbyist artist myself, I was encouraged to try out most of the technologies stated above. There I experimented with DALL-E 2 and Midjourney and based on my experience, I prefer Midjourney over DALL-E.
Even though AI art portrays a major technical feat in machine creativity, and computer vision, replicating real living artists’ work without their consent can be considered a highly unethical and irresponsible use of AI. Generative AI has caused a significant turnover in 2022 and will continue to change the course of history. If the AI-art catastrophe keeps spreading further creative competitions will begin to lose their value and artists’ careers will be at high risk.
Thus responsible AI practice is relevant more than ever. And considering the corruption that AI could inflict upon the creative industries, it is reasonable to conclude that AI has already taken over.
