OpenAI Teaches DALL-E Text-to-Image AI to Paint Beyond the Frame
OpenAI has taught its text-to-image engine DALL-E a new trick enabling the AI to viualize what might lie beyond the frame of classic and original art. The new Outpainting feature processes an existing artwork and extrapolates what might be there if the frame were bigger. For instance, Johannes Vermeer’s “Girl with a Pearl Earring” is apparently in a large, if rather cluttered, kitchen and workshop, with the synthetic elements mimicking the painter’s style and use of light to make it seem like a natural environment for the Dutch masterpiece.
Outpainting operates much like the existing DALL-E model. The AI interprets words as a prompt to synthetically generate an image. In this case, that means grafting new visuals to existing content. The difference is that the feature takes pains to imitate the style and design of what’s already on the virtual canvas, unless ordered otherwise by the user.
“Today we’re introducing Outpainting, a new feature which helps users extend their creativity by continuing an image beyond its original borders — adding visual elements in the same style, or taking a story in new directions — simply by using a natural language description,” OpenAI explained in announcing the new feature. “DALL·E’s Edit feature already enables changes within a generated or uploaded image — a capability known as Inpainting. Now, with Outpainting, users can extend the original image, creating large-scale images in any aspect ratio.”
Outpainting isn’t necessarily perfect the first time around. In the Vermeer example, artist August Kamp only widened the scene a piece at a time based on prompts provided by Kamp. Sometimes, they he needed to ask DALL-E to do it again, refining his prompts in tandem with the synthetic media production. You can see the process in the time-lapse video below.
Those interested in trying Outpainting can sign up for the DALL-E beta, which OpenAI claims has more than a million users playing with their first month’s 50 free image generations, followed by 15 additional images per month for free or 115 for $15.
Follow @voicebotai Follow @erichschwartz
Lightricks Adds AI Text-to-Image Generator to Photo Editing Mobile Apps