If you’ve been on Twitter recently, you’ve definitely seen a slew of AI-generated photos springing up all over your feed like strange, algorithmic visions. These images were created with the help of Dream, a new software that allows anybody to make “AI-powered paintings” by just entering a short description of what they wish to see. It’s strange, often unsettling stuff – and a lot of fun.
The resultant work has a distinct style characterized by whirling swirls and disordered things. The true brilliance, though, is that no matter what you input, the app will produce something aesthetically appealing (at least until we get too used to these toys) and that fits your prompt in a variety of unexpectedly appropriate ways.
A short search on Twitter yields lots more instances, but really, you should experiment with the app yourself to have a better understanding of it. (If nothing else, the photos it creates are just the appropriate size to use as a customized phone background.)
This kind of AI-generated artwork is not new, but it is getting more sophisticated and widely available. Text-to-image models have already been used in research-oriented systems like as DALL-E and VQGAN+CLIP, as well as more specialized commercial initiatives such as Artbreeder (which is particularly good at creating portraits of fictional beings and people). The AI art field has grown in recent years because to technologies like these, with practitioners generating anything from realistic Roman emperors to limitless waifus.
The Dream app goes above and above in terms of speed, quality, and accessibility. It’s made by a Canadian company called Wombo and is accessible on iOS, Android, and the web. Previously, the startup created an AI-powered tool that allows you to input in static photos to generate lip-synced versions of memeable music. It’s unclear what drives Dream (we’ve reached out to Wombo for clarification), but a lot of AI art tech is open-source, which implies the company likely relied on previous work to develop the app.
Dream’s accessibility implies that it is also being put to innovative purposes. It has been used for both viral games and more focused projects. For example, you can type in the title of your PhD thesis and share the results. In one incredible Twitter thread, writer and cartoonist Ursula Vernon (also known as T. Kingfisher) published a little comic she created using Dream. The people in the comic are hand-drawn, while the backdrops are created by AI, with the strange, changing aspect of the graphics explained by the setting: a dream library supervised by Thoth, the Egyptian god of writing.
Dream, despite its evident limitations, offers a look into the future of synthetic or AI-generated media. For technology advocates, the promise of the technology is one of unlimited variation. They claim that in the future, games, comics, films, and novels will be created on the fly in response to our every prompt and desire. And, although we’re still a long way from such media matching the quality of human production, restricted, hybrid applications will arrive sooner than you think – emerging as if from a dream.