Crayion Craiyon (DALL-E Mini): all about the AI ​​image generator that is setting the web on fire

Dall-E Mini is an artificial intelligence that generates images from texts. The result is often very strange, and this new tool ignites social networks. Find out everything you need to know.

The encounter between art and artificial intelligence often produces results that are as strange as they are fascinating. In 2016, actor Thomas Middlditch created a surprise by directing a short film based on a script written by an algorithm.

It exists today many tools to create artistic images, photo montages, music or texts thanks to artificial intelligence. Most are based on GAN-like neural networks.

Since its launch, Dall-E Mini attracts attention on social networks. For good reason, this artificial intelligence tool generates images that are sometimes funny and unusual, sometimes frightening…

As an art generation tool, Dall-E Mini create images from a short sentence input by the user. In a few seconds, the algorithm is based on the written description to generate a visual.

What is Dall-E Mini?

The Dall-E Mini artificial intelligence creates images from text entered by the user. This AI was created in July 2021 by programmer Boris Dayma, as part of a competition organized by Google and the AI ​​community Hugging Face.

This free tool uses artificial intelligence to generate images from text entered by the user. Most of the images created by DALL-E Mini are abstract and absurd, although it does occasionally produce compelling results.

Contrary to what its name suggests, Dall-E Mini is not derived from OpenAI’s Dall-E artificial intelligence. This is a much simpler and minimalist tooland much less dangerous.

At the request of OpenAI, Boris Dayma has just rename your Craiyon tool to avoid any confusion. the new website displays ads, and the developer also plans to launch a paid premium version of its image generator.

How does Dall-E Mini work?

Anyone can write a text and click on the “launch” button. The artificial intelligence Dall-E Mini produces an image in the form of a 3×3 grid containing 9 images.

This AI was trained from “ unfiltered data from the internet “. In total, almost 30 million tagged images were used for this training.

The training data were compiled by Dayma from multiple public image collections available on the web, including one put online by OpenAI.

Images produced by Dall-E Mini may be very successful or completely unsuccessful. According to its creator, the AI ​​is better at abstract paintings than portraits. Indeed, if his landscape paintings are often dreamlike and captivating, the portraits are rather hideous.

The faces are often distorted and tinged with a strange light. Photos may also be discolored, or body parts may be missing. In reality, the AI ​​produces these strange images, because it has no no idea how objects behave in the real world.

However, thanks to Machine Learning, this AI model will continue to improve over time. However, given its viral popularity on the web, the training promises to be particularly intensive.

How to use Dall-E Mini?

To use Dall-E Mini, nothing could be simpler. Just surrender on the website Hugging Face for help a computer, smartphone or tablet.

Enter your text in the field at the top of the page, and click on “Run”. Wait a few moments for the text to be processed by the AI.

A stopwatch indicates how long your request has been loaded. In general, it takes between one and two minutes.

For each request, Dall-E Mini produces nine images. You can click on any one to expand it, and right-click to save it.

If you are not satisfied with the result, you can click again on “Run” to run another query with the same text. The results are different each time.

What should I do if I get a “Too Much Traffic” error?

Since its launch, Dall-E Mini has been a victim of its own success. Of the thousands of Internet users flock on the Hugging Face website which hosts the tool, even though this site is managed by a single person.

Obviously, the platform is not not designed for such traffic. This is the reason why many attempts result in an error indicating that there is “too much traffic”.

If you are facing this error, no need to refresh the page. This will only add more pressure on the server.

Just wait a few seconds, then press the “Run” button again. Repeat the operation until the processing works and the tool generates images.

Wordalle: a game mixing Wordle and Dall-E Mini

slab and mini wordalle

Wordalle is an open-source program available on the webcombining the Wordle game acquired by the New York Times in January 2022 and images generated by Dall-E Mini. The object of the game is to guess which word or phrase the AI ​​used to produce its strange visuals.

A player enters a sentence, Dall-E Mini creates the image, and the other players have to guess the original text. Of course, this game is as difficult as it is crazy.

Note that the website does not collect personal data. Player attempts are stored in memory as part of an undisclosed “research project”, but information such as IP or email address is not stored.

The dangers of AI-created content

While Dall-E Mini can be fun to use, it also features a dark side. Its creators warn of a risk of ” reinforcement or exacerbation of societal biases “.

This artificial intelligence has no not been trained on explicit content, and it is designed to block certain keywords. However, it evolves over the requests it processes and is therefore directly influenced by each text entered by Internet users.

However, some Internet users have shared images produced from the texts evoking war crimes, school shootings or the attack on the World Trade Center.

Using artificial intelligence, it is now possible to falsify images for manipulation or misinformation: Deep Fakes.

Already in 2019, experts were worried about the explosion in the number of Deep Fake videos. In seven months he had increased by 84%.

The California was the first state to take strict measures by making the distribution of Deep Fakes illegal, especially when involving politicians. However, these contents can be very harmful even for private individuals.

In May 2022, a Deep Fake version of Vitali Klitschko, the mayor of kyiv, capital of Ukraine, began circulating on the internet. More than four European mayors would have spoken on video call with this false version. A Ukrainian President Zelenskyy’s Deep Fake was also created.

Deep Fakes are potentially very dangerous for society. Fortunately, Dall-E Mini does not currently allow to create sufficiently realistic images to create the illusion…

The best images created by Dall-E Mini

On all social networks, Internet users happily share the images generated by Dall-E Mini, adding captions written to create memes.

This tool has become a real meme machine, widely used by the community. In total, nearly 50,000 photos are created every day with the website.

Among the most shared Dall-E Mini images on the networks, we can quote Gordon Ramsay eating a Big Mac, Kermit the frog painted by Edvard Munch, Darth Vader playing the guitar, Mario and Sonic at gay pride, or Bruce Willis enjoying a yogurt.

We want to thank the author of this post for this amazing content

Crayion Craiyon (DALL-E Mini): all about the AI ​​image generator that is setting the web on fire

Discover our social media profiles , as well as the other related pages