Yokai relies on AI photo generation to appeal to general audiences and influencers

For less than five euros, Yokai offers everyone to receive in a few hours five photos of you in the universe or in the style of your choice, produced by a system based on generative artificial intelligence techniques. This French company was born in 2019 from the association of Matthieu Bizien and Olivier Zetlers and has just launched an iOS application.

The two men had already collaborated on a first project dedicated to automatic copyright protection. Serving more than 5,000 authors, his start-up analyzed millions of web pages to predict which ones were infringing in order to remove them. “A passionate engineer, Matthieu told me about generative artificial intelligence and reminded me of my first professional experience in an advertising agency. We did photo sessions, multiplying the shots and involving a huge amount of post-production work”, remembers Olivier Zetlers.

A stable diffusion based tool

To develop their solution, Yokai uses Stable Diffusion. “This is one of the top three models, it was trained from an open source dataset.” Young Shoot has developed its technology so that, from a few individual images, a person finds himself injected into the desired environment, generated by artificial intelligence. The photographic universe can also place the subject in the series War of Thrones than at Woodstock in 1969.

“The success of the series of events wednesday adams follows the movement of the core of fashion, these unitary styles of clothing passed down by generation Z. We are currently creating a collection in this core of fashion so that people can get an extremely realistic picture of themselves in 80s punk or dark style Academy of Wednesday.” Yokai also plans to release new themes like California’s famous Burning Man festival.

The copyright issue

To place a Yokai user at the center of this American event, generative artificial intelligence will use Stable Diffusion’s macro model and billions of photos from the festival. “Theoretically, the great models of generative AI were built from a corpus of royalty-free images. To respect copyright, the AI ​​must not copy without authorization, so the question is: when the AI ​​produces an image, it is all or part of this photo taken from an existing image? Basically no!”, replies Olivier Zetlers.

Concerned with being transparent on the subject, he recognizes that 1 to 1.5% of the images generated are liable to copy elements. His defense: “These are AI accidents that the labs involved are working on. You can’t blame AI for copying when it learns from the existing one.” These systems just copy, that’s the basic principle of their operation. They mimic the works they were trained in and are especially capable of producing hybrid results from many different sources.

The co-founder believes that generative AI mimics the operation of an artist who is inspired by the work of another to make a painting in a similar but by no means identical genre. “The AI ​​decomposes shapes into complex equations to generalize its understanding. The notion of copying is destroyed in its process.” It should be noted, however, that the question of the application of intellectual property in the scope of image generation has not yet been decided on a legal level.

From the general public to professionals

Yokai puts itself first at the service of individuals with the aim of gradually reducing the cost of its service to become free. “Our system requires powerful, expensive, power-hungry graphics cards,” justifies the co-founder. Generation Z is targeted first, but the start-up intends to open up to other markets in 2023, starting with influencers, professional photographic models and all those audiences that multiply shots and must quickly generate new images.

The co-founders will also look to the B2B market, first with photo printing services that can incorporate the Yokai solution to develop their next generation photo book offering. “We will also approach fashion and beauty brands to offer users the opportunity to generate photos of themselves in the brand’s environment and thus immerse the customer in its universe”.

In the long term, probably in 2024, Yokai will continue its rollout reaching advertising agencies. “Our solution still does not meet the expectations of this market in terms of flexibility and functionalities.” What about professional photographers? “The photographer who brings an artistic dimension will be preserved, he brings his eye to compose the final image, AI will help him.”

protect from deviations

At the same time, Yokai wants to protect itself from potential abuse and abuse of its technology. Some can, in fact, project an image from someone else’s photo. “There’s little risk because the editorialized environments and the imagery possibilities generated remain framed.” Additionally, Yokai makes a point of having a human verify each photo before sending it out to users. “The system is based on images taken decades ago, potentially carrying racist or sexist content from another era. It is programmed to avoid its presence, but the human verifies that the algorithm worked.”

Finally, to prevent an image thus generated from being presented as a photo taken in a real context, Yokai has implemented a content identification system capable of recognizing its own pixels and thus specifying whether the photo was generated by it.

Selected for you


Leave a Comment