Creating amazing images using text is getting popular day by day. Today, you can imagine the most random sentence in your head, and putting it in a text-to-image generator will give you astonishing results! This can mean a lot of things for the AI industry, creative industry, illustrators, and everyone in between. DALL-E 2 is one such tool created by Open AI that many people are eagerly waiting to explore. Here is everything you need to know about Open AI’s tool, DALL-E 2:
What is DALL-E 2?
DALL-E 2 is an AI system created by the AI company Open AI and comes with the ability to create images using a text description. What is exceptional about this tool is that it can create entirely new images by using the knowledge it works on. The name DALL-E resembles the 2008 Disney movie WALL-E and the artist Salvador Dali. It is the second iteration of the previously launched DALL-E and one can say that this is an improved and updated version. In fact, DALL-E 2 is outright better than the original DALL-E in every way.
For example, if you input the text, “a white paper with zig-zag lines on a chair”, DALL-E 2 will display several image options of the exact input from the scratch.
The Infinite World of DALL-E 2
When you explain it so simply, people can often miss out on how amazing this tool is. DALL-E 2 can not only create images but also create short videos. You can take a look at this short video posted by a Twitter user. The tool can transform a Victorian house into a house that looks like it is built in modern times and vice versa. DALL-E 2 can diminish timelines by connecting something that was prominent in the past with today and also adding something that can potentially be in our world in the future.
You can also give the tool-specific inputs related to photography and create images that will incorporate this effect. To click these pictures manually is a different thing and generating them through AI is something extremely mindblowing. To existing images, you can also use it to add a specific element. For example, you can type “add a table” and the tool will give you the desired results.
How Does DALL-E 2 Work?
If you want to understand it in an easy way, this text-to-image generator uses a diffusion model that AI Researchers referred to as a neural network.
It also uses the CLIP model, developed by OpenAI that matches the caption or text to the images. It has been trained using the millions and millions of text-image pairs that are available all over the internet.
So, if the tool has seen millions of images or analyzed them, it can differentiate between two objects or elements. The diffusion model helps in the latter part of combining or creating pixels and ultimately generating an image for the user.
As someone who is looking to generate images through the tool, it is very simple to use it. But right now the question is when will everyone be able to use it?
How to access DALL-E 2?
This tool is not available for everyone to use and you first have to sign up for a waitlist to access it. And although summing it up in one line doesn’t sound so bad, the CEO of OpenAI, Sam Altman still needs some figuring out to grant access to everyone on the waitlist.
After looking at the potential that DALL-E 2 has, the number of waitlist signups seems fair. On May 18th, 2022, the company updated that after spending the last month researching and testing the limitations, it is planning to onboard at least 1000 people every week from the waitlist.
If you still can’t wait to explore your creativity and want to get a hands-on experience of the potential of DALL-E, then you can take a look at DALL-E mini. This is a really toned-down version of DALL-E 2. This tool is still under development but it can give you an idea of how DALL-E 2 works.
The Limitations of DALL-E 2
Although it is true that with such text-to-image generators, you can have a lot of fun with your creativity. But it’s a thin line. If the AI tool has learned from many images all over the internet, it has also picked up on information that has the potential of causing harm to the world.
There is a reason why text-to-image generators including DALL-E only show and promote images that look cute and represent everything good in the world. Creating tools that explore everything that AI can do has the potential to revolutionize the world but only if there is a way to filter out the dangerous stuff.
“As we work to understand and address the biases that DALL·E has inherited from its training data, we’ve asked early users not to share photorealistic generations that include faces and to flag problematic generations.”
DALL-E 2 is also open for only a limited number of users because of the same reason. One of OpenAI’s blog posts read, “As we work to understand and address the biases that DALL·E has inherited from its training data, we’ve asked early users not to share photorealistic generations that include faces and to flag problematic generations. We believe this has been effective in limiting potential harm, and we plan to continue the practice in the current phase.”
The introduction of tools like these can give us an idea of how much artificial intelligence has progressed. It will be truly amazing to see how artists and illustrators can use DALL-E 2 to explore their potential and bring their imagination to reality.
Featured Image Credits: OpenAI Website