The fever of generative artificial intelligences did not start with ChatGPT. A year before the arrival of this conversational application, OpenAI surprised the Internet community with Dall-E, a tool that was able to imagine any scene described in natural language. In April 2022, months before making ChatGPT public, it presented, in fact, the second version, which finally produced realistic results at an acceptable resolution.

Its arrival triggered interest in the use of artificial intelligence in creative tasks and helped drive the development of alternatives such as Midjourney or Stable Diffusion, engines that today are capable of imagining any scene and drawing it with any style. Now, OpenAI has the third Dall-E version ready and promises to once again raise the bar for generative artificial intelligences by solving some of the usual failures of this type of creation engines.

A notoriously difficult problem for Midjourney or Dall-E 2, for example, is people's hands. They tend to draw them with more or less fingers than normal and not always in anatomically correct positions. The same goes for teeth. There are usually too many. They are small details that allow you to quickly identify an image created with artificial intelligence.

Another thing that chokes these apps are the texts in an image, often displayed only as a meaningless set of lines and dots. Dall-E 3, says OpenAI, is capable of generating readable text in images and creates natural-looking hands.

The tool will be available to users of the ChatGPT+ subscription service from October and the company will also include it in enterprise plans for access to its APIs. This is not an additional app. To use Dall-E 3 it will be enough to ask ChatGPT to draw the scene.


The company says it has taken some extra precautions to avoid copyright infringement issues. For example, you can't be asked to paint a scene in the same style as a living artist.

This is one of the biggest problems that the organization has faced after ChatGPT achieved enormous popularity. Several researchers have discovered that to train the GPT-3 language model (the engine that gives life to ChatGPT), the company used books and articles that were not part of the public domain. Several authors have sued the company on the grounds that it is profiting from their work without adequate compensation.

The same happened with the images used to train Dall-E 2 and other creation engines. Illustrations and photos posted in communities and art forums such as DevianArt were used without consulting the creators and artists behind them.

The company will now offer creators a tool so that their creations, especially those published on the network, are not used as part of the training of future artificial intelligences.

  • Artificial intelligence
  • ChatGPT