Thanks to AI, Jellysmack star transformed YouTubers into rich YouTubers


This American unicorn relies on artificial intelligence founded by three French people to optimize the visibility and revenue of video makers on social networks.

Founded in 2016 in the United States by three French individuals, Jelismac relies on AI to optimize the presence and revenue of video content creators on social networks. In 2021, the company stopped raising confidential funds from SoftBank, Partech and Highland Europe. An operation that brought it into the circle of the Unicorn, with a very strong ambition for international development. With 100 manufacturers signing in early 2021, the company is growing to over 500 this year. Among them are PewDiePie, people with 110 million subscribers on YouTube, Brad Mondo, Bailey Sarian or Quet. Jellysmack has more than 1,000 employees worldwide, including 700 in France.

“We mainly identify YouTube channel creators who are starting work, and we suggest adapting their content for distribution on Facebook, Instagram, Snapchat, TikTok, Twitter …”, Jellysmack’s CTO, Thierry Bedos summarized. . The company has been using AI since the designer identification phase. Capitalizing on image recognition and automated language processing, particularly the Kibert model, it uses clustering algorithms to classify videos and identify high-potential individuals based on their audience mobility.

Capital as a service

From there, Jellysmack offers the manufacturers a financing. How does he calculate his bet? It uses a time-series-based neural network to predict revenue from their catalog over the next five years. Based on these assumptions, he signed a check for them. Thierry Bedos adds “that allows them to invest in equipment, to hire them to accelerate development”.

“We use computer vision to convert YouTube’s 16/9 format to Facebook’s square format or TickTock’s 9/16 format.”

Then comes the stage of adapting videos to targeted social networks. A stage where AI clearly occupies a key place. It optimizes the primary content according to the features of each platform, in terms of player, duration, usage … “We use Computer Vision to move YouTube’s 16/9 format to Facebook’s square format or 9. Tiktok’s / 16. To automate the blurring of visual elements, Thierry Bedos explained.

Another challenge: moving from a 20-minute Youtube video to a 3 to 4 minute Tiktok video. Here, AI intervenes to identify passages that reflect a powerful emotion: a character who expresses himself emphatically, who cries or who laughs, for example. In the final edit, JellySmack editors will capitalize on all this preparatory work to save valuable time.

Optimize video content

JellySmack creates several variants of the same video to quickly test users in a small sample in A / B testing mode. Since this population is small, machine learning intervenes to extrapolate the results according to the learning model of the society. Models that are fed into a mega database created by the publisher It combines more than one billion public videos collected from over 10 million YouTube channels and their respective public viewers statistics. Attached is the personal information of the start-up’s own customers. Finally, based on the performance of the selected video, JellySmack is paid to share the advertising revenue with the producer.

For the future, Jellysmack is working on two ways to develop products. First, the company wants to make its video editing chain smaller by relying more on AI. Finally, it plans to provide its clients with recommendations and advice on video types with high potential in their respective departments.

Leave a Comment