Generative AI: once upon a time the image revolution

Can artificial intelligence be creative? If the algorithms are not artists, their ability to “imagine” images from scratch from a simple written request is stunning.

Is creativity a sign of intelligence? Can you be free under duress? So many philosophical questions that could apply to generative AI. In recent months, we have been hearing more and more about this new concept of image generation using artificial intelligence (AI). Perhaps you have seen the names of DALL-E (GPT-3)MidJourney or Stable Diffusion, to name only the best known tools.

They are able to create in a handful of seconds a painting, a photo or a drawing based on a simple descriptive sentence. It can be “an astronaut on a horse”, but also “a colorful 3D rendering of a happy robot artist painting a painting with stripes”, or “panoramic view of a chair in the style of pop art on forest background. Is it therefore a work of art? “I don’t believe that the machine creates anythingsays Monty Barlow, director of machine learning at Cambridge Consultants. She takes an area that humans are extremely good at and she finds the mechanics behind it. But we still need to give it an initial input. »

Generative AI: creative programs under constraints

Concretely, it is an artificial neural network capable of solving complex problems, in this case the creation of images, following learning on the basis of millions of images and associated legends. Algorithms gradually learn artistic styles (drawing like Van Gogh), the characteristics of living beings (representing a baby, a tapir), emotions (joy, fear, etc.) or even places (at the beach, at the foot of the Eiffel Tower…).

Then, the power of computers and artificial intelligence algorithms make it possible to create original images starting from constraints (the instructions given by a request formalized by a person) and adapting them to new situations. Each time you submit a request, also called a “prompt”, the system will create a new visual, fairly close to the previous one or very different. It is this random variable in the algorithm that is the closest one can imagine to human creativity.

“If you have a specific idea in mind, you have to describe everything: what subject, what angle of the camera, what artistic filter, do you want a centered or offset framing, what texture, do you imagine an image with a lot of color or neutral, do you want a photo or a painting, etc. »

The art of talking to artificial intelligence or happy accidents?

These “happy accidents” are what appeal to Geoffrey Dorne, a 37-year-old freelance designer who has embarked on the “prompt art” since the beginning of 2022 and shared on his Twitter account AI-generated images. Its concept? Via the hashtag #IAFP, he creates sticky images with news headlines from the AFP news agency. “I love testing the latest tech innovations. But I never test in a vacuum. On Twitter, I had noticed that AFP news was regularly unillustrated or with ‘pretext’ illustrations. I wanted to see if AI could solve this problem”he explains.

But how do you talk to a Generative AI tool? If we can imagine at the beginning that it is a bit like learning to formulate a search in Google, the prompts can be much more complex. “I learned the basics of sign language a few years ago and I realize that it helps me a lot to create prompts because you really have to be very descriptiveemphasizes the designer. To evoke a door handle, the sign language will describe its use (a metal bar on which one presses to open a door). It’s the same with Generative AI. If you have a specific idea in mind, you have to describe everything: what subject, what camera angle, what artistic filter, do you want a centered or offset framing, what texture, do you imagine an image with a lot of color or neutral, do you want a photo or a painting, etc. »

The art of speed… and all the business around it

Geoffrey Dorne works a lot with the AI ​​MidJourney – particularly trained for an artistic rendering when Dall-E is better in realism – which is accessible on a Discord server, a discussion platform. “The advantage is that we can follow all the requests submitted by others in real time. This allows you to see how others are doing to create their prompts and then launch », explains Geoffrey Dorne. But it is also possible to be helped. The website Phraser.tech offers a free tool that guides step by step the creation of images so as not to forget (almost) anything in the description.

PromptBase goes even further. This is a paid site where anyone can come and sell their “prompts”, whether for creating images or writing texts (what we call AI Writing and which is based on the same principles). The Marketplace shows an example of the results that can be obtained using one of these prompts (each generation of images will be different). If you like this one, you only need to pay an average of 2 to 5 dollars to use it. You then discover the detail of the prompt (a long text that goes into detail with sometimes quite technical information). You will then be able to use this description yourself by adapting it to your needs.

The prompt created by Phraser to generate this image in Midjourney is “Multicolor sketch of a journalist writing a story by Blender Guru, made of a fluffy dandelion texture, medium shot, with a sense of inspiration and of fun. »

The thorny issue of copyright

Could generative AI one day replace artists? It seems unlikely. “Anyone can make a prompt, it’s true, but illustrators or even photographers are on the lookout for thousands of small details (unique look at a subject, angle, context, narration, point to make, etc. ). We can also have a political and social commitment that is reflected in our creations, which an AI will never have”notes the designer.

There remains the question of copyright. Geoffrey Dorne credits and sells the creations imagined by IA thanks to his prompts, but he makes it a point of honor to always rework them afterwards to improve the result and adapt it more to his taste. “It’s a bit like the first samples of hip-hop music. There, the AI ​​will dig into millions or even billions of images. And it’s my prompt that will give life to a new creation. » He is not alone: ​​there are more and more of these AI creations on established image banks, such as Shutterstock. They are identified as such and sold in the same way as a photo or a drawing.

“Under French law, only a human being can be the author of a work and create an original work. However, in the case of Prompt Art like Dall-E, SnowPixel, MidJourney… we consider that there is no human intervention in the very creation of the image. And the robot has no rights. »

Betty Jeulin

Digital Lawyer

Is it really legal? Betty Jeulin, digital law lawyer, explains that there is still a great vagueness in terms of legislation. “Under French law, only a human being can be the author of a work and create an original work. However, in the case of Prompt Art like Dall-E, SnowPixel, MidJourney… we consider that there is no human intervention in the very creation of the image. And the robot has no rights. » It is therefore impossible to qualify these creations as “works of the mind” and we then fall into a kind of legal vacuum.

A legal wild west… that will not last

“The copyright does not belong to the person who created the prompt. Nor can we really attribute paternity to the people who trained the algorithm. Pending clearer legislation, reference should be made to the conditions of use of AI image-generating tools. Often it says the image can be used by crediting the tool, which retains ownership. On the other hand, it is not not possible to trade »explains Betty Jeulin.

The European Union leaves for the time being each country to legislate according to its will in the matter. In France, a law on artificial intelligence is expected for the beginning of 2023, but it will focus on critical risks, for example in health and transport. Copyrights will not be included. But the societies for the collective management of these rights, such as the Society of People of Letters (SGDL) or the Sacem for music, are in full reflection… and could soon rule each differently on the question. One thing is certain, we are moving towards the obligation to clearly specify when a “work” has been generated by artificial intelligence.

Finally, there is the question of the rights of the artists whose images have been used to train the algorithms. As difficult as it may seem to attribute any authorship to an artist for a detail in an image, it is understandable that representatives of Picasso or Dalí threaten prosecution of creations made specifically “in the style of”. “If an artist’s “paw” is easily recognizable, then we enter the domain of counterfeiting. Copying a work and profiting from it is already a case punishable by law.warns Betty Jeulin.

The DeviantArt platform, which allows artists to exhibit online, has just announced a new option: an “opt-out” for artists to prohibit generative AIs from using their creations to train their algorithm. But how the isolate among hundreds of millions of visuals already submitted and ensure that an AI will “unlearn”? A real headache.



We would love to give thanks to the writer of this post for this incredible web content

Generative AI: once upon a time the image revolution


Visit our social media accounts and other pages related to themhttps://www.ai-magazine.com/related-pages/