Strange but realistic camera concept created by DALL E 2 AI

Create off-the-wall camera images from scratch … Thanks for the artificial intelligence. This is the real challenge that Matthew Stern has set in a recent video. Thanks for a new AI system, called Slab 2It is possible to create an image from a few words, change the style of a photo or a table, or add or remove elements.

A fun – and amazing – way to show off How AI understands our world, And how it combines the elements together. Discovery.

DALL · E 2: AI at the service of your imagination

Have you ever seen a picture of a koala? dunk Basketball? Or astronauts on horseback to the moon? Or Mona Lisa with Iroquois crest? No? It’s quite normal, since This image does not exist.

© OpenAI – DALL · E 2

Laboratory researcher to unveil new creative horizons for ordinary people Open AISan Francisco-based, has launched a new artificial intelligence in 2021, named Slab. In today’s 2nd edition, it allows To get much more realistic resultsFor both photographers and artwork.

“Our hope is that DALL · E 2 will allow people to express themselves creatively. DALL · E 2 helps us Advanced AI systems can understand how we see and understand the worldWhich is essential to our mission to create an AI that benefits humanity, ”said the project’s designers.

On paper, its operation is very simple: OKWrite a few word descriptions. And AI takes care of image creation Which, of course, made the video an overnight sensation. A point that is very reminiscent of Google’s Imagine project, which we presented to you a few days ago.

Camera in Star Wars, Back to the Future or Lord of the Rings

Matthew Stern was a photographer and YouTuber Opportunity to test DALL · E 2 in preview. Taking advantage of the power of this AI, he was able to simply design one Original-fake camera series Bigger than life, which adopts the stylistic code of the most famous movies and series.

We can make a discovery like this Medium format cases like R2-D2, C3-PO or Darth Vader. Or a camera of the same color as Marty McFly’s jacket in the movie. Back to the future.

Similarly, we can find Angular (even oppressed) case design like a solarInside The Lord of the RingsAs well as a roliflex decorated in the style of one ring.

Many creative examples that illustrate what AI can create with a few words. That’s it, Perhaps Matthew had to go through trial and errorRefine the results obtained at each iteration.

Nevertheless, this exhibition is particularly interesting because it shows that AI connects and integrates different concepts. Image Descending Compatible And realistic from a human point of view.

Of course, Nothing stops you from bringing these ideas to life through Photoshop, But the operation is long enough – and requires a very advanced knowledge of software Here, the interest is to expose this possibility to the maximum number in just a few seconds.

DALL E 2: An extra trained AI for ultra-realistic results

To achieve such a result, OpenAI called on researchers Acquiring deep knowledge. In practice, AI analyzes a database of millions of so-called “labeled” images.

Each image is captioned in a few words, AI is able to “understand” what Visual presentsAnd to weave logical links between different elements (e.g. rider and its mount).

Beyond the semantic perspective, AI also learns to recognize the style of an image or table. This way the user can ask AI Transform the Mona Lisa into an Impressionist paintingFor example.

Especially strong, this instance “Image from text” Also gives the possibility to edit an existing image. In the example photo quoted, we can thus replace the poodle sitting on the sofa with a cat.

Open AI researchers say that “DALL · E 2 has learned the relationship between images and the text used to describe them.” “It uses a process called ‘diffusion’, which starts with a pattern of random dots and gradually converts that pattern into an image because it recognizes certain aspects of that image.”

Thus, the system begins by creating a “mush of dots”, which combines them into a realistic whole – and in high definition.

What are the limitations for this new artificial intelligence?

Like any artificial intelligence, DALL E 2 encounters various limitations. On the one hand, if the source image caption is incorrect, AI will make some errors.

Also, the relationship between objects, ideas and styles is sometimes very complex. As Matthew Stern shows, sometimes the desired question needs to be refined a few times, Because AI does not immediately understand what the user has in mind.

More broadly, DALL · E 2 (and its competitors) pose a A number of moral issues. Of a Hand, there is a negligible risk Strengthen or reproduce some stereotypes – What project designers say they want to avoid at any cost.

On the other hand, DALL · E 2 was driven by a Deliberately selecting images excluding any content of a violent or sexual nature, So that AI cannot create such content. Similarly AI has been “Forbidden” so that it does not make a human face – and especially public figures.

In fact, DALL · E 2 (and its competitors) have a greater risk of using AI Make something Deep fakeThese fake photos (and videos) are sometimes very realistic.

Since then, OpenAI researchers have chosen a radical solution. As a result, DALL · E 2 is only available to a select number of selected users. In Episode 2, AI will be available by invitation – but with a Dual-monitoring system (Human and automatic) to avoid any kind of abuse.

DALL E 2: A powerful tool in the service of the creative mind?

Some image manipulation projects can be a cause for concern – and often just so. Its spread in a context Fake newsThe advent of tools that allow you to create “driven” photos (or videos) in just a few seconds can be more of a real concern than nature.

Nonetheless, the initiative of OpenAI researchers shows how artificial intelligence can be relevant. From a creative point of view, DALL · E 2 makes it possible to give life to more or less bizarre ideasThe only limitation is with our imagination – which Matthew Stern comes to illustrate brilliantly.

Similarly, AI can be used in very relevant ways to transform our scribbles into photorealistic decorations, restore old videos or bring to life portraits of our ancestors.

From a technical point of view, DALL · E 2 shows us how AI perceives and interprets our world. An important point for the development of artificial intelligence in the service of humanity.

For more information about DALL · E 2, visit the project website. And find other videos of Matthew Stern on his YouTube channel.

Leave a Comment