AI text-to-image generators have already given us an idea of their potential. From DALL-E to Craiyon, people have used these artificial intelligence tools to create memes, predict how the end of the world will look, and a lot more.
The new outpainting update by OpenAI, the creators of DALL-E is showing the world it will keep getting better. DALL-E, already had the option to edit or modify something from an existing image, called inpainting edit but it only created images of 1024 × 1024 px.
With Outpainting, users can now extend any image to create an image with a wide angle of any aspect ratio. The tool considers the textures, shadows, and reflections of the image to maintain its context.
In the announcement of this update, OpenAI included a time-lapse of Johannes Vermeer’s portrait “Girl with a Pearl Earring” and how the artist August Kamp used small sections to expand and portray the background of this painting. It also gives us an idea of how the tool can expand with the help of limited data and maintain the painting style of the artist. These outpainting generations are created by giving user prompts.
This increases the capability of the tool and users can further get creative with this feature. While you can use it to expand what exists in the image like the above-mentioned artist has done or also use your imagination to create something entirely different. That’s the beauty of AI.
Similar features are available on other text-to-image generators but what will make one of them stand out is making it easy and offering an update that requires minimum manual inputs. Open AI’s DALL-E 2 is now available to a million users. It has also laid out certain guidelines to avoid any kind of misuse of the tool. In this test version, users are given fifty free image generations in the first month and fifteen from the following month. Users can also pay extra if they want to generate more generations.
This could be a limitation for users who want to explore outpainting as generating one 1024×1024 px additional outpainted area will also take one generation from the users. So, you can do the math for guiding the AI to create an entire image with this feature like the artist created in the time-lapse.
Featured Image Credits: Open AI/ DALL-E 2/ Artist August Kamp