The initial release of DALL·E was revealed in January 2021 by OpenAI. This AI system could generate images from simple text descriptions.
For example, you type in "avocado armchair," and DALL·E will generate just that.
AI Generated Avocado ArmchairsCredit: OpenAI
DALL·E 2 takes this concept even further by providing higher-resolution images and greater text-to-image comprehension. New capabilities were added, such as inpainting, which can replace parts of an image with new text-generated art.
The new iteration can even start with a completed image and create new images of different variations, angles, and art styles.
Unrestricted Creativity with DALL·E 2
DALL·E was created by training a neural network on images and their text descriptions. Through this deep learning technique, the system not only understands individual objects but also understands the relationship between them.
So when you ask DALL·E 2 for an image of a koala bear riding a motorcycle, it delivers.
AI generated imagesCredit: OpenAI
Images are created with a process called diffusion. This starts with a pattern of random dots and gradually alters that pattern toward an image when it recognizes specific aspects of the image you are trying to generate.
There are limitations, however.
The AI system is only as good as its training. If the system is trained to see a plane as a car, it will generate the wrong image. There can also be gaps in its knowledge, such as for specific species. While it can understand what a monkey is, it might have trouble figuring out how to generate a howler monkey.
While the research is focused on developing the framework to see how much the system can understand our world, it delivers a robust tool for creatives to spark their imagination.
The wrong kind of howler monkeyCredit: OpenAI
Creating with DALL·E 2
While DALL·E 2 is still in the research phase, creatives who may get access now or in the future can implement the system in several fantastic ways.
Imagine storyboarding a short film with heavy VFX story elements. Punch in your description to allow DALL·E 2 to spark your imagination or do the heavy lifting.
We can also expand this to other creative disciplines. Writers, concept artists, and fabricators can generate new creature concepts, vehicles, or worlds so they can focus more time on creating.
'Terminator 2'Credit: Tri-Star Pictures
The Future of Creative AI
For most creatives, this technology is merely a thought experiment, at least for now. But it opens a window into a future that is all about pushing the boundaries of your creativity. Check out the possibilities over on the OpenAI blog.
At best, this kind of tech supercharges your imagination, but at worst, it may be the start of our very own Skynet.
Whatever the outcome, the tools that creatives may get access to in the future will feel more like Star Trek than Terminator, and that’s a good thing, in our book.
Source: OpenAI
