DALL-E2, OpenAI’s improved version of DALL-E (a nod to “WALL-E, the 2008 animated movie about an autonomous robot, and Salvador Dalí, the surrealist painter) is a neural network that can turn natural language into stunning images, thanks to advanced deep learning techniques. DALL-E2 can create realistic, never-seen-before visuals from a simple text prompt. An example might be a tyrannosaurus playing the accordion while riding a unicycle on the moon. If you tweak the prompt with additional detail, it will generate a whole new array of images. The possibilities are unlimited because the AI is built upon approximately 40,000 years-worth of artwork dating all the way back to cave drawings.
In addition to unique image generation, it can also replace things in existing images, such as inserting a different person or object into a photo. It can also take an image as an input and create many variations with different styles and angles.
It’s trained on a massive collection of images with label text descriptions. Most importantly, it understands how different objects and images relate to each other – like how a teddy bear relates to a skateboard relates to Santa Monica.
How is an image generated? A decoder or transformer uses a process called “diffusion”. It starts with a pattern of random pixels which are gradually moved around until they satisfy the text input. To make high quality images, it generates 512 possible outcomes, then uses a neural network called Clip to predict human aesthetic judgments by ranking the results and rejecting the ones that don’t look good.
DALL-E2 has important implications for Tomestic as explained by Founder and CEO, Poet De’Medici, “There are so many different things you can do with an image prompt that can provide a unique Organic Living Book (OLB) experience. Each book will be a ‘living’ entity, unique to its owner, by virtue of its birthdate, location, and content. AI will generate images through voice command and text, such as producing a different cover of the OLB each time it’s used. Whatever content you upload will change the book, with the possibility of offering thousands of different book experiences in one device.”
“We can customize a classic book such as Alice in Wonderland to the uniqueness of the OLB. The OLB will know the reader’s likes and dislikes. So, a hybrid text prompt may generate Alice in Florida or Alice in Alaska. Each user’s imagery will be completely unique and reflect his or her preferences.”
Tomestic has big plans that will include all the capabilities of AI. Please sign up for our newsletter on the ‘Home’ page so we can keep you informed.