DALL-E 2 is an AI system that can create realistic images and art from a description in natural language. It is one of the amazing programming advancements that have captured the world’s imagination. We thought you might like to know a little more about it.
DALL-E 2 was created by AI research lab OpenAI and publicly launched in July 2022. It’s a successor to their original DALL-E model. It uses a deep learning technique called “diffusion models” to generate highly realistic and creative images from simple text descriptions. Users can give it prompts like “an armchair in the shape of an avocado” and it will generate a novel image.
The images it creates are very high-resolution (1024×1024 pixels) and the quality has improved a lot compared to the original DALL-E. The images have finer details, better textures, more realistic faces, etc. DALL-E 2 has been trained on huge datasets of image-text pairs from the internet, allowing it to make associations between visual concepts. This helps it “imagine” new combinations of objects, attributes, and styles.
In user testing, people found DALL-E 2’s images to be more photorealistic and higher quality than other AI image generators. The variety and creativity of the images is also impressive. DALL-E 2 represents a big leap in what AI can do in terms of creating original, human-like visual art and content. Obviously, the applications are far-reaching in terms of digital art, design, content creation, etc.
And, as you can well imagine, even though OpenAI has implemented usage policies to prevent abuse, there is on-going debate around ethics, especially related to creating misleading or harmful content.
DALL-E 2 still has elements that are in beta, however I was able to start using the service by going to: https://labs.openai.com and creating an account. Then I had to buy credits to start making images. But the cost is very reasonable. I bought 15 credits for about $15. I then asked it to produce a photo of a woodchuck eating a flower in our back yard and it produced the image shown. You can see that the resolution and depth of the image are excellent – comparable to what you might expect from a stock photo house. And DALL-E 2 gave me a bunch of angles, positions and other variations to choose from. All that ate about 5 of my 15 credits. (I wish the woodchuck did as little damage!)
As we are all learning with AI, the devil is, of course, in the “prompts.” Learning what to ask your AI resource, and how to build on the responses to get the best response. With art, that conversation is further complicated by the need for judgement, training and talent – what we like to call the Human Intelligenceelement.
DALL-E 2 is not likely to replace our professionally trained and talented art directors anytime soon, but it sure is fun to play with. For simple PowerPoint graphics and home projects it makes us all feel like we are Ansel Adams.
So, give it a try. We are likely in the middle of the most significant revolutionary technical development since the advent of the computer. The more you know at a hands-on level, the better.
Incidentally, I used the generative AI resource Claude 2 to help with information about DALL-E 2. It helped me write some of the content above but told me that DALL-E 2 was not yet available to the public. That turned out to be incorrect. Another example of the need for Human Intelligence.
Now let the fun begin. Email me an image you create with DALL-E 2.