new OpenAI creates images from words

  • 52

At the beginning of last year we told you about DALL-E, the OpenAI artificial intelligence capable of generating images from any text we wrote. The new version of this system, under the name of DALL-E 2, is now official, and the improvements over the previous one are remarkable.

DALL-E 2 is being tested in a private development environment, so we can’t ask you to draw what we want yet. However, since OpenAI they have shared multiple images of its operationand the results are spectacular.

“Create an image of an astronaut riding a unicorn”

Tg Image 606640420

DALL-E 2 is the revision of the artificial intelligence model that we saw last year, and it is more spectacular than ever. OpenAI defines it as “a new AI system that can create realistic images and art from a natural language description.” It is currently in a closed testing phasebut there is numerous examples about its operation.

Like the previous model, this AI takes advantage of the potential of GPT-3 to generate its designs, through a complex system in which first an image is generated given a text caption and, subsequently, a second image conditioned to the first is generated.

Tg Image 1683487426

Tg Image 1428692595

In addition to creating images from scratch, it is capable of make image edits, as well as recreations of them from their own model. The results are much more fine-tuned than those seen in the last generation.

Tg Image 214836201

“Painting of a fox sitting in a field at sunrise in the style of Claude Monet.”

We have interviewed GPT-3 about what life will be like in 2030 and this is what he has told us about Mars and the future (and he says his name is Luis)

They explain from OpenAI that DALL-E 2 has better learned the relationships between images and the text used to describe them, now being able to generate images with a resolution four times higher.

To prevent misuse of AI, DALL-E 2 cannot generate images considered violent or explicit. The generation of photorealistic images is also limited to avoid the recreation of real people’s faces.

More information | OpenAI

At the beginning of last year we told you about DALL-E, the OpenAI artificial intelligence capable of generating images from…

At the beginning of last year we told you about DALL-E, the OpenAI artificial intelligence capable of generating images from…

Leave a Reply

Your email address will not be published.