Data map (from:OpenAI）
On Wednesday, the company added a new feature called “outpainting” that allows users to expand the boundaries of images.
According to OpenAI engineer David Schnurr, it can expand relevant content based on existing image and text cues.
Limited by processing power, DALL-E can only expand existing images for the time being, and cannot generate higher-resolution content.
But users want different aspect ratios and want to be able to expand concept creations into larger images.
FROM-E TO Fashion TUtorial – Karen X Cheng （via）
As an example of AI’s technological capabilities, the codename DALL-E also combines the names of Pixar’s “WALL-E” robot and the surrealist painter Salvador Dalí.
To achieve this, OpenAI trained on 650 million images with textual descriptions. DALL-E can mix and match various creative elements based on text prompts.
Interested friends can submit a registration application first, and after passing the waitlist, they can use this service to generate 60 images for free each month. Product manager Joanne Jang revealed that millions of people have signed up for DALL-E so far.
Judging from the various content shared on Twitter, the creativity of netizens is also quite unrestrained. Other serious uses include storyboarding for movies, illustrating children’s books, and exploring concept art such as video games.