The outwardly simple app, which generates nine images in response to any typed text prompt, was launched almost a year ago by an independent developer. But after some recent improvements and a few viral tweets, its ability to roughly outline all sorts of surreal, funny, and even nightmarish visions suddenly became meme magic. See its renderings of “Thanos is looking for his mother at Walmart,“”drunk guys without upper body wander around Mordor,“”CCTV camera footage of Darth Vader breakdancing,” and “a hamster Godzilla in a sombrero attacking Tokyo. “
As more people created and shared DALL-E Mini photos Twitter and Save your, and several new users arrived, so Hugging Face saw its servers overwhelmed by traffic. “Our engineers did not sleep the first night,” Clément Delangue, CEO of Hugging Face, said in a video call from his Miami home. “It is really difficult to operate these models on a large scale; they have to fix everything. “In recent weeks, the DALL-E Mini has served about 50,000 images a day.
DALL-E Mini’s viral moment not only heralds a new way of making memes. It also provides an early look at what can happen when AI tools that make images to order become widely available, and a reminder of the uncertainty of their possible impact. Algorithms that generate custom photographs and artwork can transform art and help companies with marketing, but they can also have the power to manipulate and mislead. A warning on the DALL-E Mini Web site warns that it may “aggravate or aggravate societal imbalances” or “generate images that contain stereotypes about minority groups.”
DALL-E Mini was inspired by a more powerful AI imaging tool called DALL-E (a portmanteau by Salvador Dali and WALL-E), unveiled by AI research firm OpenAI in January 2021. DALL-E is more powerful, but is not open available due to concern that it will be abused.
It has become common for breakthroughs in AI research to be quickly replicated elsewhere, often within months, and DALL-E was no exception. Boris Dayma, a machine learning consultant based in Houston, Texas, says he was fascinated by the original DALL-E research paper. Although OpenAI did not release any code, it was able to merge the first version of the DALL-E Mini at a hackathon hosted by Hugging Face and Google in July 2021. The first version produced low-quality images that were often difficult to recognize, but Dayma has kept improving it ever since. Last week, he renamed his project to Craiyon, after OpenAI requested that he change his name to avoid confusion with the original DALL-E project. The new site shows ads, and Dayma is also planning a premium version of its image generator.
DALL-E Mini images have a markedly foreign appearance. Objects are often distorted and smeared, and people appear with faces or body parts missing or damaged. But it’s usually possible to recognize what it’s trying to depict, and it’s often fun to compare AI’s sometimes unreleased output to the original prompt.