Perhaps you recognise this: you have searched for an image, but it did not exist. Or the image you had in mind was very different from what was available online. The artificial intelligence programme DALL-E offers a solution to this. Thanks to this programme it is possible to describe an image in single words. Then DALL-E will put together an image for you. Examples of searches? A red cat in the style of Picasso, a realistic picture of an elephant on a ladder or a group portrait of people with ants' heads. Get as crazy as you want! Let your imagination run free...
How does DALL-E work?
DALL-E is designed by OpenAI, a research laboratory for artificial intelligence (AI). From search words, DALL-E generates images. The model was released in 2021 and is capable of combining realistic objects and imaginative subjects. DALL-E can create images in different styles; from realistic photos to paintings and emojis. It is special and extraordinary to actually see your thoughts in front of you.
Funny fact: The name is a fusion of Pixar's animated character WALL-E and the Spanish painter Salvador Dalí.
The future of DALL-E
You might think: away with drawing or editing programmes like Photoshop! Wouldn't you? That is not yet the case. OpenAI has not yet released the source code for the model. Recently DALL-E 2 was released, in which the quality and resolution of generated images has been improved considerably. Also, images can now be easily edited, something which was not possible before at all. This 2nd version of the image generator takes the 'text to image'-skill to an even higher level! Will DALL-E be widely used by visual designers, interior designers and other creative people in the future? Perhaps...
Side note of the model
The incredible possibilities of DALL-E make it dangerous. Realistic images can be generated that may be offensive, discriminatory or hateful. Fortunately, OpenAI recognises this and has removed certain content from the training data. The lab has also minimised DALL-E's exposure to these kinds of concepts. Finally, OpenAI has ensured that DALL-E cannot generate realistic pictures of real people (including famous people).
The quality of DALL-E tests > test 1
We have used a free available version, namely DALL-E Mini, to test the model. With these keywords "rainbow carrot jumping laughing on a mountain" we get several images. You see a laughing carrot (or several) with sometimes a rainbow in the background, on top of a mountain and a somewhat scary laughing face. Mission accomplished! Admittedly, the images are not really usable because of the poor quality.
The quality of DALL-E tests > test 2
Time for a more difficult task, which is more specific and more extensive. When we type in the search words "a dancing black cat with a purple hat and fairy wings on the disco floor", we see exactly that: a magical black cat. Mission accomplished! Although very accurate, and matching the keywords, the quality is still a bit fuzzy or childish.
The quality of DALL-E tests > test 3
This one we like very much: "emoji picasso style dog". There is enough variation between the pictures and DALL-E has kept the deformed style of Picasso. The different shades of blue, such as blue-green, also come back. We are starting to get the hang of it now...
Become a fan of DALL-E
After reading this article, are you eager to try out DALL-E for yourself? Go to the free version and type in the craziest keywords! We love the latest technical gadgets. We also follow AI developments closely and think about how we can make these trends our own. Would you like to run an eye-catching campaign or place a special post? We are happy to think along with you about the implementation. Give us a call!
