An adorable dachshund wearing diving goggles under the water of a swimming pool, a poodle sailing on a flamingo buoy, maple syrup dripping on pancakes… These videos lasting a few seconds are more real than life. yet never been filmed. They are the work of Veo 2. A few days ago, Google unveiled this model allowing you to generate a video of a few seconds from a simple description.
With its AI inspired by a worm, Liquid AI seduces investors
A few days before that, OpenAI made accessible – not in the European Union however – Sora, its own video model announced last March. Pika, a video generation app, also announced new features. This end of the year is tinged with excitement around these AI video models.
229 euros per month for ChatGPT Pro, OpenAI's crazy bet
Booming video start-ups
But beyond the bluff effect, what is the market for these synthetic images? Although in its infancy, the technology is already attracting the attention of many sectors. As proof, Runway, a pioneering start-up in the field, should end 2024 with $84 million in turnover, reports The Information. That figure is expected to triple to $265 million by the end of 2025, according to company projections. Its growth would then be one of the most significant in the sector. Runway, like most AI players, offers a subscription. About $75 per month in his case.
The type of customers who purchase these subscriptions is varied. “ AI-generated videos are very popular with individuals, people who simply want to create videos for their children or for social networks.explains Thibaud Zamora, creator of the Fictions.AI studio, the first in France dedicated entirely to films produced by AI.
On the professional side, “We find influencers who use these tools to have more content, visibility and therefore to monetize this traffic. The advertising industry is interested in it. And it is also arriving in the film industry, for the moment not for finished products, but to produce models which are used to present a project”adds the entrepreneur. Thibaud Zamora's studio has produced around ten short films and an interactive game, largely made with generative artificial intelligence. He has also been approached by brands to produce their advertising.
Advertising with AI, a new standard
Advertising is in fact the ideal “early adopter” for technology. “Even if the quality is not always exceptional, the short lifespan of advertisements makes this aspect less important”remarks Thibaud Zamora. Recently, major brands have displayed their “AI” spots, like Coca-Cola with its famous Christmas ad, McDonald Japan, Toys R Us, Motorola, etc.
“All the competitors of these brands will inevitably get involved. In 2025, using AI for advertising will be the new norm »believes Frédéric Raillard.
His agency, Fred & Farid, is one of the first to bet big on AI. At the beginning of 2024, she launched an independent studio called [Ai]magination, all of which is produced using technology. Ultimately, Frédéric Raillard believes that his agency's spin-off will far exceed the original.
« 80% of advertising content will be made with AI. Faced with ever-increasing needs for video content, the industry cannot resist technology that is better, faster, less expensive and more durable than existing technology. »he believes.
The studio already has around thirty projects for 20 brands and 5 agencies. Not all videos produced use text-to-video models, which directly produce a video from text. Sometimes, you still have to go through the generation of images and then animate them with a so-called image-to-video model.
Financial and time savings
The savings in terms of cost and time compared to traditional production methods would already be significant, according to the various sector players interviewed. “A subscription for an artificial intelligence model costs roughly the price of a person’s taxi to a shoot”compares Frédéric Raillard. For Pernod Ricard, its agency delivered visuals and videos for 8 advertising campaigns in a few days. “If we had to shoot everything, it would have taken 6 months. »
Argil.ai, a French start-up specializing in the creation of hyper-realistic avatar videos generated from a short video of a real person, has made lower costs its main argument. “We go from a budget of 200 euros to pay a creator of a one-minute video, compared to one dollar with our service”thus praises Laodis Menard, founder of the young company which has just raised 3.9 million euros in seed funding.
“AI is also used to do things that we could never have done in any other way”notes Arthur Kannas, of the Heaven agency. In a commercial video for Motorola a few months ago, the agency created, using Luma (a video generator), a fashion show where the models' clothes ended up forming the brand's logo.
“With traditional techniques, this would have been impossible, or it would have taken 8 months and a lot of special effects…”he judges.
The manager, however, remains less optimistic than his colleague about the large-scale deployment of the technology. For him, AI must remain a tool used when it makes sense. “For Motorola, which is a brand on the comeback, the idea was to put it back at the heart of discussions around AI”he explains. He also notes that ethical and legal questions, particularly around the rights of images used by model suppliers, are holding back many brands from launching.
Poor quality videos
“Initially, the generation of videos by AI will be limited to videos with low added value, but produced in large quantities, such as promotional ads found on social networks, or short product presentations on e-commerce platforms »judges the communicator.
Amazon, TikTok and Meta have in fact integrated tools to generate video advertisements in just a few clicks, announcing a flood of synthetic promotional content. For its part, Argil ensures that a significant portion of its customers use its services to produce so-called “UGC” (User Generated Content) videos.
These videos, often presented in the form of customer testimonials, abound on social networks. Today, they are mainly produced by freelance content creators. Thanks to Argil's technology, these videos can be mass generated at low cost, offering several variations from a single initial sequence.
Furthermore, if the savings in money and time made possible by these models are difficult to question, the quality of the renderings is more so. The Coca-Cola spot seems very bland compared to the brand's previous Christmas special productions, which were high-end.
New models continue to improve but still remain imperfect. One of the major limitations of Sora, the OpenAI model, is its misunderstanding of the physical rules and the inconsistencies of the images generated. From one second to the next, one element can blend into another, or a face transform erratically. Veo 2 promises better performance, but for now it is not available to the public.
Fiction films soon?
Thibaud Zamora brushes this problem aside. “The video will follow the same path as the image. Two years ago, it took me 25 minutes to get an image with AI, and it took at least 1,000 minutes to get a good one. Today, in a few seconds and only four tries, we obtain a quality image. » According to him, in the coming months, we will see more and more beautiful short formats generated by AI: advertisements, trailers, etc.
“And in three to five years, it will be possible to make fiction films”he projects.
“Given the speed at which this technology is advancing, quality will soon no longer be an issue”agrees Frédéric Raillard. « On the other hand, we will always need and want to see real humans on screen. Furthermore, for certain brands, for example sports, which sell effort and surpassing oneself, putting synthetic beings on screen will always seem questionable. We must continue to support traditional productions, because we will need both »he judges. But if we follow its projections (80% of advertising productions generated by AI), seeing real humans in advertisements could become a rare privilege.
