OpenAI introduces DALL·E 2

DALL-E 2: anything in any style imaginable

OpenAI introduces DALL·E 2: anything in any style imaginable - examples
Source <a href=httpscdnopenaicompapersdall e 2pdf>OpenAI<a>

Do you sometimes lack imagination when people describe things to you? Do you need design inspiration? Or do you wonder how all the new funny, surreal memes on Twitter and Instagram come about? The photo collages that look amazingly real but aren’t? That’s exactly what the latest invention from OpenAI, an Artificial Intelligence research laboratory, is making possible right now. They already launched the first version of a text-to-image system in January 2021 and now OpenAI introduces DALL·E 2: anything in any style imaginable.

DALL-E, named after the Spanish surrealist artist Salvador Dali and the computer-animated Disney movie WALL-E, is capable of visualizing a countless range of surreal ideas and concepts you might have. In fact, OpenAI is a powerful and versatile creative tool that uses artificial intelligence to create all imaginable images based on comments and descriptions of your idea. No matter which absurd non-real description you enter into the AI system, you get images of the exact explanation as a result. 

 Sam Altmann, the CEO of OpenAI, announced that DALL-E 2 not only creates higher quality images than the first version but also changes various options of already existing images and then shows different styles, angles, and variations of your description within 10 seconds. Whether you’re a designer looking for a chair that looks like an avocado, want to create funny memes of famous people, or want to create surreal images of animals acting like people, the text-to-image generator creates possible options for the imagination within seconds. 

Sam Altman also tweeted that DALL-E 2 performs much better with “longer and more detailed” descriptions, implying that simpler sentences perform worse because they are too general for the text-to-image generator.


OpenAI introduces DALL·E 2: – new functions:
  1. in-painting

Insert an already existing picture to DALL-E 2 and it will realistically edit and retouch it just the way you want it to be! 

  1. interpolation

OpenAI also found a way to transform one image into another and combine and blend the two, with DALL-E. Boris Dayma, the founder of DALL-E mini, for example, mixed the Starry Night, an already existing painting of Van Gough, with an actual picture of two dogs and within seconds the image-generating system combined these completely different pictures into a deceptively real-looking new image.

interpolation example - Starry Night, an already existing painting of Van Gough, mixed with a picture of two dogs
Source <a href=httpscdnopenaicompapersdall e 2pdf>OpenAI<a>



OpenAI introduces DALL·E 2 – three main outcomes:
  1. The text-to-image generator helps people to express themselves visually in ways they haven’t been able to previously. OpenAI created with DALL-E a new standard and revolutionized not only the design branch but is also an illustration of how creative people and AI systems may collaborate to create new stunning things.
  2. AI-generated images reveal a lot about how well the system already understands humans or whether it is just repeating what it has been told to do. DALL-E 2 uses deep learning to understand the details and relationships between individual objects to then connect them in the right way and create AI art.
  3.  DALL-E aims to improve people’s understanding of how advanced artificial intelligence systems perceive and understand the human world. 


These are just some of the reasons why the DALL-E 2 waiting list, for a free trial, is endlessly long. People are fascinated by the latest invention of the OpenAI research laboratory and want to try out the text-to-image generator themselves and create DALL-E 2 examples. 


OpenAI introduces DALL·E 2 – limitations & disadvantages:

Nevertheless, DALL-E 2 is still not a perfect, final version of artificial intelligence and has its limitations. Artificial intelligence is only as smart as its programmers. 

  1. If disinformation was introduced, like a wrong naming for certain objects, DALL-E will also deliver the wrong result without knowing it’s incorrect.
  2. If the text-to-image generator has never learned a term, it does not know it, but at most tries to derive it from other terms. Again, deep learning is used, yet the knowledge of DALL-E 2 is still limited.
  3. The third point is probably the most problematic in the subject of artificial intelligence art. This is the critical social point. OpenAI programmed DALL-E to create images using biases and stereotypes. Thus, if the term nurse is entered, only women are displayed, just as with pilots, for example, only images of men appear, although there are also employees of the opposite gender in both occupational fields. Also, OpenAI has not yet succeeded in introducing a filter for explicit content, bullying or harassment and one can therefore also generate very offensive images with DALL-E.


The DALL-E 2 release date is still not official. But Burst Creative Group is just like a million other people already on the DALL-E 2 waiting list. We are eagerly waiting to see and try what possibilities and opportunities this AI revolution will bring to us as a company and will of course show you a full review on our news page as soon as we had the chance to try it.