+++lead-in-text

New products often come with disclaimers, but in April the [artificial company issued an unusual warning when it announced a [new service called DALL-E The system can generate vivid and realistic photos, paintings, and illustrations in response to a line of text or an uploaded image. One part of OpenAI’s release notes that “the model may increase the efficiency of performing some tasks like photo editing or production of stock photography, which could displace jobs of designers, photographers, models, editors, and artists.”

+++

So far, that hasn’t come to pass. People who have been granted early access to DALL-E have found that it elevates human creativity rather than making it obsolete. [Benjamin Von an artist who creates installations and sculptures, says it has, in fact, increased his productivity. “DALL-E is a wonderful tool for someone like me who cannot draw,” says [Von who uses the tool to explore ideas that could later be built into physical works of art. “Rather than needing to sketch out concepts, I can simply generate them through different prompt phrases.”

DALL-E is one of a raft of new AI tools for generating images. [Aza an artist and designer, [used open source software to generate a music for the musician Zia Cora that was shown at the [TED in April. The project helped convince him that image-generating AI will lead to an explosion of creativity that permanently changes humanity’s visual environment. “Anything that can have a visual will have one,” he says, potentially upending people’s intuition for judging how much time or effort was expended on a project. “Suddenly we have this tool that makes what was hard to imagine and visualize easy to make exist.”

It's too early to know how such a transformative technology will ultimately affect illustrators, photographers, and other creatives. But at this point, the idea that artistic AI tools will displace workers from creative jobs—in the way that people sometimes describe robots replacing factory workers—appears to be an oversimplification. Even for industrial robots, which perform relatively simple, repetitive tasks, the evidence is mixed. [Some economic suggest that the adoption of robots by companies results in lower employment and lower wages overall, but there is also evidence that in certain settings [robots increase job way too much doom and gloom in the art community,” where some people too readily assume machines can replace human creative work, says [Noah a digital artist who posts YouTube tutorials on using AI tools. Bradley believes the impact of software like DALL-E will be similar to the effect of smartphones on photography—making visual creativity more accessible without replacing professionals. Creating powerful, usable images still requires a lot of careful tweaking after something is first generated, he says. “There’s a lot of complexity to creating art that machines are not ready for [Keep
##### Search our [artificial intelligence and discover stories by sector, tech, company, and more.
+++

+++

The first version of [announced in January was a landmark for computer-generated art. It showed that machine-learning algorithms fed many thousands of images as training data could reproduce and recombine features from those existing images in novel, coherent, and aesthetically pleasing ways.

A year later, DALL-E 2 markedly improved the quality of images that can be produced. It can also reliably adopt different artistic styles, and can produce images that are more photorealistic. Want a studio-quality photograph of a Shiba Inu dog wearing a beret and black turtleneck? [Just type that in and A steampunk illustration of a castle in the clouds? [No Or a 19th-century-style painting of a group of women signing the Declaration of Independence? [Great people experimenting with DALL-E and similar AI tools describe them less as a replacement than as a new kind of artistic assistant or muse. “It's like talking to an alien entity,” says [David R a photographer, writer, and English teacher in Japan who has been using DALL-E for the past two weeks. “It is trying to understand a text prompt and communicate back to us what it sees, and it just kind of squirms in this amazing way and produces things that you really don't expect."

Munson likens DALL-E’s responses to his prompts to the weird or surprising logical connections made by the young children he teaches. He asked the program to create an “anthropomorphic pot roast holding a Bible,” imagining it would produce something like a pot of stew with eyes, but he got something quite different. “It made these weird, lumpy meat-men,” he says. Munson also used DALL-E to recreate a vivid memory from his childhood, of watching television news of the fatal [Space Shuttle accident in 1986.



[#image: R Munson used an AI tool called DALL-E 2 to recreate his memory of seeing a TV news report about the 1986 Space Shuttle *Challenger* disaster.|||

The new version of DALL-E is just one example of a new category of powerful image-generation tools. Google recently announced two, in May, and in June. Several open source projects have also created image generators, such as [Craiyon, which went viral last after people began using it to post memes on social media.

New companies have sprung up to commercialize artistic AI tools. A website and app called can generate images in a variety of styles in response to a text prompt or an existing image, and it sells prints or of the results. an independent research lab that has made its technology available to beta testers, can turn text prompts into vivid, sometimes abstract illustrations.

David Holz, the founder of Midjourney and previously CTO of [Leap Motion, a 3D computer interface company, does not see his tool competing with “We're focused on exploring the essence of imagination,” he says. “Imagination is used for many things, sometimes art, but more often simply reflection and play. We wouldn't call what we make AI-art, as the AI doesn't make anything on its own. It has no will, no agency.”

Midjourney runs a Discord where beta testers can submit a prompt for the company’s algorithm to work with. Many people testing the service are artists, Holz says. “They feel broadly empowered and optimistic about using the technology as part of their workflow.”

DALL-E and many other AI art tools are built on recent advances in [machine that have enabled algorithms that process text or images to operate at much greater scale and accuracy. A few years ago, researchers found a way to feed huge volumes of text scraped from novels and the internet into these algorithms, allowing them to capture statistical patterns of text. After that training, the system [could generate highly convincing when given a starting sentence.

Similar AI models have since proven adept at capturing and recreating patterns from other data, including audio and digital images—the basis of DALL-E. But these image-generation systems lack any real understanding of the world and can produce images that are glitchy or nonsensical. And because they replicate the web-sourced images they were trained on, they can reflect societal biases—for example, always rendering doctors as male and flight attendants as female. There is also the potential that such programs could be used to generate fake photographs that are used to spread misinformation.

OpenAI has these risks and says it has implemented measures to prevent DALL-E from being used to create objectionable or misleading imagery. Those include preventing the system from generating images in response to certain words, and restricting the generation of celebrity faces.

The errors and glitches of AI image generators can themselves feel like an artistic tool. a less-capable clone of the original DALL-E previously named DALL-E Mini, went viral last month after users discovered the fun in providing it with surreal, farcical, or unnerving text prompts. [One art describes the limitations of the AI behind Caiyon as yielding an “*online grotesque*”—bizarre or disturbing fusions drawn from the zeitgeist of the internet. Popular examples include “[muscular “[gaming or “[Death star gender clever prompts are at least half the fun,” says [Aaron a principal scientist at Adobe Research and an affiliate professor at the University of Washington who studies computational art. He says Craiyon and other image-generation tools are enabling new forms of exploration, something inherent to creativity. And he compares text-to-image tools to a kind of conceptual art similar to that of [Sol or [John where the idea behind a piece can be its most important component.

Perhaps the biggest change that AI image generators will bring is dramatically expanding the number of people able to generate and experiment with art and illustration. “Optimistically, you might say this is revolutionary in communication,” says [Tom White](https://drib.net/), an artist based in New Zealand whose work explores [artificial Even those who are not artistically inclined could use such tools to generate and share creative images, White says, something people are already doing with Craiyon memes. “That may change how we relate to each other.”

White, who’s artwork includes [abstract images carefully crafted to fool common image-recognition programs](http://ngines), says he enjoys testing DALL-E 2 to try and reveal aspects of the images in its training data, and what restrictions have been placed on the system to prevent creation of offensive images. Over time, he begins to see a kind of “personality” in the missteps a particular system makes.

White suspects that tools like DALL-E 2 may become far more powerful and interesting as it becomes possible to interact with them in different ways. The only way to refine an image DALL-E produces currently is to rewrite the prompt or crop the image and use it as the prompt for a new set of ideas. White believes that it won’t be long before people using creative AI tools will be able to ask for specific adjustments to an image. “Dall-E is not the end of the road,” White says.

*Additional reporting from Tom Simonite.*

Source: Wired

Powered by NewsAPI.org