Tuesday, October 18, Los Angeles. The Adobe Max event keynote is about to begin. Several thousand spectators gather in the huge Microsoft Theater with a simple question in mind: are we going to talk about generative artificial intelligence (AI)? For the past few months, the art world has had eyes only for this technology. Its pioneers are called Dall·E 2, Stable Diffusion or Midjourney. These sites make it possible to digitally create anything and everything. A bit like Adobe, via Photoshop, one of the best-known software published by the company. With the difference that they only need a few words of description, a command (translated into prompt in English) so that works are generated for free. As if by magic.
Their limit seems to be only that of our imagination. A fox propelled into a painting by Claude Monet, legends like Freddy Mercury or a Michael Jackson aged as if they were still alive, or a simple field of poppies flooded with sunlight… These visions of the mind appear on a screen in just a few seconds, versus hours of work on a drawing board or on Adobe and many other software.
In August, an American artist even won an art competition in Colorado, USA, using the Midjourney platform. Of course, without the jury realizing anything. The information circled the globe because it delivered a simple message: AI, already used sporadically for technical improvements, has not only tricked the human mind – deepfakes already do – but it has also shaped what is unanimously recognized as “beautiful”.
Limited offer. 2 months for 1€ without commitment
In the huge auditorium, the suspense is finally quickly lifted by an Adobe manager, David Wadhwani. The company will also offer “magical” alternatives in its creation tools, in particular for designing new fonts. The announcement receives its share of clamor and immediately raises new questions. Are art professionals going to be outdated by technology and by a wave of neo-artists for whom all technical barriers have now been lifted? Will creative tools like Adobe’s become useless? “We see a time coming when not everyone has the time or the intention to learn Photoshop,” admits Maria Yap, in charge of AI at the company. Dall·E 2, a time reserved for a handful of insiders, has become accessible to everyone and already attracts more than 1.5 million users, while generating 2 million images per day.
In search of the spark
Among the designers, graphic designers or even photographers encountered at the Adobe Max show, the vast majority have tested generative AI. Often, at first, for fun. “I entered the words ‘alien’ and ‘ice’ on Dall·E 2 to see what came out,” smiles Estella, an artist specializing in virtual reality, a helmet on her head. Lionel Koretzky, a French photographer, has also taken the plunge for the “unknown and exciting” side that emerges from these platforms. “When you are an artist, you are in perpetual quest for the artistic accident, this spark that you did not expect, and these tools can provoke it”, he confides. He then inserted a real work “brief” on Midjourney, an order from a client, to see what came out of it. The result, not uninteresting, nevertheless required adjustments to be presentable.
An Eiffel tower in the “studio Ghibli” style made with the Midjourney tool.
Maxime Recoquillé / L’Express
Because the tool is not as magical as it claims. Designer in the creation company Monotype, also present in Los Angeles, Terrance Weinzierl believes that generative AI models mainly create “empty and artificial” works. Nothing of a nature to compete with human vision and sensibility. Or quite simply talent. The creations made using Dall·E and its cousins are far from all successful. With imprecise commands, large defects are easily visible in faces or textures. The computer does not always understand what is said to it. And then there are bad ideas. A Twitter account, Weird Dall-E Generations (weird Dall·E works), followed by a million people, compiles these totally failed essays: a cat in a coal mine, actor The Rock hugging dictator Kim Jong-un… For now, a lot of software refuses to show nudity or violence which can be part of a respectable creative process. Internet is what it is.
This takes nothing away from the revolution that these tools represent. For artists of all kinds. “AI is a primer like photographic prints were for painters at the start of the last century. Like penciling, drafting,” observes information and communication science researcher Olivier Ertzscheid. But also for creative businesses. “We will quickly value those who master the quickbecause these orders to create illustrations, if they are well mastered, will be a huge time saver”, says Frédéric Cavazza, consultant specializing in marketing and advertising. These bits of sentence are already the basis of a budding business : Count from 3 to 4 dollars for those who are sure to generate interesting results, female heroic fantasy characters or pretty clothes.
An image of a dish designed on Midjourney using the words “Italian gastronomy” in particular.
Maxime Recoquillé / L’Express
Keep your hand on the wheel
“We will soon be able to ask him to explore 30, 130 or 130,000 different options of the same creative idea. Before letting us choose a small number to work on them personally, to deepen them to finally find the best options to present. customers. It’s a real superpower,” enthuses Scott Belsky, chief product officer at Adobe, which cautiously refers to a “co-pilot”. Implied: of its own tools. A vision that gets along: Adobe is one of the biggest sellers of creative software in the world, a company valued at nearly 150 billion dollars that does not want to see its business go up in smoke. The company also has reason to be cautious on the subject. Scott Belsky compares the rise of generative AI to that of the self-driving car. For safety, “it is still necessary to keep one hand on the steering wheel”.
First for ethical issues. Dall·E 2 is privately owned. Its tool is based on a language model that allows the machine to recognize text and its meaning (via machine learning). It uses one of the most advanced in the world, OpenAI’s GPT-3, released in May 2020. As well as a neural network called Clip, still developed by OpenAI, which very roughly operates the link between text and an image within a huge corpus. The bigger it is, the more varied the results. But this platform, like many others, today guards this jealously guarded database, which raises legal questions. “If we cannot listen to it, we accept the idea that there will be serious road trips, such as racist or sexist biases”, points out Olivier Ertzscheid. Which would not be a first in the history of AI. Building your own artistic bases could be the future of generative AI, but this formula still seems distant. The technology is not within everyone’s reach and the costs in terms of operation and storage – the machine requires a large amount of information to learn – can quickly become significant.
Legal issues, particularly related to copyright, are the other big question mark for this technology, the progress of which makes it possible to tackle not only images, but also code, sound, video and even 3D. Should artists be paid if their work is used to inspire others? And how do you know? Lionel Koretzky says he tested an order by affixing the name of the Japanese artist Yayoi Kusama, who became famous for her polka dot works. An image faithful to the style of the artist came out. The Express has also tested styles, such as that of the famous Ghibli studios, which the machine has imitated without the slightest problem. Dall·E 2 has taken some precautions on this subject, for example refusing to take into account “Philippe Starck” from the name of the French creator, in order to protect his works. A restrictive decision to avoid any legal trouble. While after all, the pastiche, thanks to the AI, seems to have good times ahead of it.
opinions
The chronicle of Christophe Donner
Chronic
The chronicle of Christophe Donner
Chronic
