Dall · E 2, the AI ​​which generates images according to your imagination

Google imagen

Dall-E: the famous IA image generator is available for free

For the past few weeks, the Dall-E artificial intelligence software has been talking about him. And to the delight of the lovers of this technology, the company Openai, which is behind its development, has now made it accessible to all and for free.

As a reminder, Dall-E is a tool that allows you to generate images from text queries indicated by Internet users. But in recent weeks, it has not really been unanimous, especially with the artistic sphere which judges the software as an attack on creativity. A debate that has also accentuated when a work created entirely using artificial intelligence software Midjourney won the 1st prize for an art competition, as we mentioned in this article.

If previously Dall-e was only reserved for professionals registered on a long waiting list, it is now accessible to everyone freely and free. According to Openai, it would already count more than 1.5 million active users generating more than 2 million images per day.

By making it free to use, American company wishes to do everything to improve its tool and control its drifts. For example, it is impossible to generate portraits of known public and political personalities, nor nudity and violence. OPENAI also wishes to offer real and inclusive representations of the world, by limiting the racial biases which were accused of in its early days.

Each registration for Dall-E has 50 free credits, which is equivalent to 50 creations. Subsequently, 15 additional credits will be added each month and for the most impatient, it is also possible to buy packs of 115 credits for 15 euros. To find out more or test Dall-E, go to the OpenAi website.com.

Share the idea on your networks

Dall · E 2, the AI ​​which generates images according to your imagination

Appointed Dall · E 2, according to the film Wall-E by Pixar and the surrealist painter Salvador Dalí, this neural network is being developed by the Research Laboratory in Open AI Artificial Intelligence, co-founded by Elon Musk.

Have you ever seen a koala playing basketball?

Thanks to deep learning, Dall · E 2 not only includes individual objects, but he also learns the relationships between objects. For example, he is able to deduce what a koala pulling in the basket would look like, although he has never done this before.

Dall · E 2 can also edit and retouch photos realistically from a simple description in natural language. It can fill or replace part of an image with images generated by the AI ​​which blend perfectly in the original. It can even start from an image and then produce variations with different angles and styles.

While the first version of this technology made results a little blurred in a cartoon style, Dall · E 2 can produce more precise images in a variety of styles with better quality and more complex backgrounds.

What are the risks?

The concerns around Dall · E 2 are the same as those mentioned for Deepfakes: the disinformation of disinformation on the Internet. Its developers have published a warning in this regard on the site: “Without sufficient safeguards, models like Dall-E 2 could be used to generate a wide range of misleading contents”.

“Legally, the questions raised by this technology are just as dizzying,” said the newspaper Next Inpact. Determine to who owns an image created by an AI is always subject to debate.

Dall · E 2 is not accessible to the general public, the algorithm is currently being tested by some chosen users – in order to better define its capacities and its limits.

Dall-e Mini

On the other hand, it is possible to do your own trials in a less successful version of the software, the mini dall, to better understand what it is. Try, it’s pretty stunning.

Here is what “a cat sleeps in an armchair”, “Swiss neutrality” and “Putin on a tank”. But still (added on 11.07), “Trump painted by Picasso”, “The robot portrait of Mark Zuckerberg” and “John Lemon”.

»»..

Google imagen

Google, develops a similar tool, named Imagen, launched on May 24. We cannot test it, but several examples created by AI have been shared on their site.