DALL-E 2 Legacy in AI Art

When OpenAI’s DALL-E 2 debuted on April 6, 2022, the idea that a computer could create relatively photorealistic images on demand based on just text descriptions caught a lot of people off guard. The launch began an innovative and tumultuous period in AI history, marked by a sense of wonder and a polarising ethical debate that reverberates in the AI space to this day. Last week, OpenAI turned off the ability for new customers to purchase generation credits for the web version of DALL-E 2, effectively killing it. From a technological point of view, it’s not too surprising that OpenAI recently began winding down support for the service. The 2-year-old image generation model was groundbreaking for its time, but it has since been surpassed by DALL-E 3’s higher level of detail, and OpenAI has recently begun rolling out DALL-E 3 editing capabilities.

But for a tight-knit group of artists and tech enthusiasts who were there at the start of DALL-E 2, the service’s sunset marks the bittersweet end of a period where AI technology briefly felt like a magical portal to boundless creativity. Illustrator Douglas Bonneville recalled the exhilaration and sense of unlimited freedom that DALL-E 2 initially brought. The arrival of DALL-E 2 was a truly mind-blowing experience, offering a glimpse into the transformative potential of AI in the realm of creativity.

Before DALL-E 2, AI image generation tech had been building in the background for some time. Since the dawn of computers with graphical displays in the 1950s, people have been creating images with them. As early as the 1960s, artists like Vera Molnar, Georg Nees, and Manfred Mohr let computers do the drawing, generativity creating artwork using algorithms. Experiments from artists like Karl Sims in the 1990s led to one of the earliest introductions of neural networks into the process. The use of AI in computer art picked up again in 2015 when Google’s DeepDream used a convolutional neural network to bring psychedelic details to existing images. Then came generators based on Transformer models, an architecture discovered in 2017 by a group of Google researchers. OpenAI’s DALL-E 1 debuted as a tech demo in early 2021, and Disco Diffusion launched later that year. Despite these precursors, DALL-E 2 arguably marked the mainstream breakout point for text-to-image generation, allowing each user to type a description of what they wanted to see and have a matching image appear before their eyes.

When OpenAI first announced DALL-E 2 in April 2022, certain corners of Twitter quickly filled with examples of surrealist artworks it generated, such as teddy bears as mad scientists and astronauts on horseback. Many people were genuinely shocked by the capability of the technology. Other examples of DALL-E 2 artwork collected in threads soon followed, all of which were flowing from OpenAI and a group of 200 handpicked beta testers.

When OpenAI began handing out those beta testing invitations, a common bond quickly spawned a small community of artists who felt like pioneers exploring the new technology together. Conceptual artist Danielle Baskin, who first received an invitation to use DALL-E 2 in March 2022, recalled the thrill of having access to a portal into infinite alternate worlds. This sense of exploration and the merging of language with visual imagination created a novel and exciting experience for the testers. Each word in a prompt functioned like part of an address pointing the generator to a different spot in the conceptual latent space, bringing unique mixtures of visual elements to the fore.

DALL-E 2 couldn’t render images perfectly, but the imperfections gave the images a dreamlike, painterly quality that the testers found attractive. It also began to impact their artwork in the physical world. The group of testers remained close on Twitter, pushing the limits of this entirely new medium together. As they explored the intersection between language and visual arts, they began to think of themselves as explorers of “latent space,” a term for the compressed multi-dimensional neural network representation of everything the AI model has absorbed.

During this period, ethical questions surrounding the technology began to surface. Some outside the initial testing group expressed concern about DALL-E 2’s ability to generate art by analysing hundreds of millions of images scraped off the Internet without consulting rights holders. Blogger Andy Baio articulated these concerns, questioning the ethics of training an AI on copyrighted creative work without permission or attribution, allowing people to generate new work in the styles of existing artists without compensating them, and charging money for a service built on the work of others. These ethical dilemmas led to polarising views on social media, with some users celebrating AI art and others condemning it.

OpenAI’s tight lid on things initially kept the controversy to a minimum. The company claimed ownership of all generations, included a built-in content filter for violence, famous people, and sex, and added a small corner watermark to each generated image. However, as other AI image models like Midjourney and Stability AI’s Stable Diffusion became publicly available, the practice of using existing artists’ names in prompting techniques caused a backlash. This issue led to a lawsuit against Stability AI and Midjourney and sparked ample protests in online artist communities.

For the early DALL-E 2 testers, the initial sense of wonder and joy in exploring AI artwork began to fade as ethical concerns and commercial hype took over. Each artist had to navigate their own ethical stance on AI-generated art. Some, like Bonneville, took a more permissive view, believing that creative work shared on the Internet should be expected to be used by fair use principles. Others, like Lapine, became uncomfortable with the technology after discovering personal medical records in the training dataset.

As the novelty of AI image generation wore off, some testers felt disillusioned with the endless onslaught of AI-created images. The community that once thrived on Twitter began to fracture, partly due to changes in the platform and the collapse of the cryptocurrency-linked NFT market, where some AI artists hoped to earn money.

Now, as DALL-E 2 is being phased out, the development of AI image generators continues with new models like DALL-E 3 and advancements in synthetic video technology. Despite the controversies, the early days of DALL-E 2 remain a significant period in the history of AI, marked by a unique blend of creativity, exploration, and ethical debate. The experiences of the early testers highlight both the potential and the challenges of integrating AI into the world of art.

for all my daily news and tips on AI, Emerging technologies at the intersection of humans, just sign up for my FREE newsletter at www.robotpigeon.be