open AIAI text-to-image generator developers DALL-E 2just announced a new feature for the app called “surpassing”. It allows users to augment existing images and artwork with AI-generated content. It’s quite exciting and greatly expands the capabilities of the tool.
DALL-E 2 is one of the most popular text to image generators available at the moment. With over a million users, it is not surprising that the content created by him seems to be everywhere. (Many other text to image generators are in closed beta, like stable diffusionare not available to the public, as Google imagesor have a much more limited range, like Craiyon).
DALL-E 2 takes a text message, like “an astronaut riding a horse in the style of Andy Warhol,” and generates nine 1024-pixel-by-1024-pixel images that illustrate it. It uses a process calleddiffusion” where he starts with randomly generated noise and then edits it to match the salient features of the message as closely as possible.
Until now, users were limited in the size and aspect ratio of what they could create with DALL-E 2. The AI program could only generate squares that were 1024 pixels by 1024 pixels; anything larger or differently shaped was out of reach. Question. It was possible to use a feature called “repaint” to modify details in existing artworks, but to create a larger canvas, different sections needed to be manually stitched together using an application like Photoshop. (For different aspect ratios, you can crop your image, but that reduces the overall resolution.)
Now with outpainting, the only limit users face, aside from content filters—are the number of credits they have. (Everyone gets 50 free spawn credits for the first month and 15 to use every month after that. Additional 115 credit blocks can be purchased for $15.) Generating an initial image requires a credit, as does each additional section painted.
Outpainting works as an extension to DALL-E 2. Users select a 1024 pixel by 1024 pixel square area where they want to extend the image and can specify additional prompts to guide the AI. For example, to add more background to the astronaut on a horse, you can change the message to “an astronaut riding a horse on the moon with stars in the background in the style of Andy Warhol.”
For each painted section, DALL-E 2 will offer up to four possibilities for users to select. If none of them work for the image, you can have it try again.
Most impressively, the top painting “takes into account the existing visual elements of the image, including shadows, reflections, and textures.” This means that any added detail “maintains the context” of the image and can actually appear to be part of a cohesive whole.
In the external painting ad for DALL-E 2, there is a time lapse showing the pearl girl by Johannes Vermeer being enlarged to about 20 times its original size. Instead of a simple portrait, it shows a young woman standing in a messy house. It’s fascinating to watch because as long as you don’t look too closely, it really does look like an extension of the original painting. The overall style and mood are perfect. It’s almost like an imaginary behind-the-scenes shot.
If you want to try painting, you’ll need to sign up for DALL-E 2. Open AI is currently operating an ongoing waiting list. If you want to register, you can do it here.