Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Meta will license technology from artificial intelligence image and video generation start-up Midjourney, as the social media group shifts towards working with third parties as its struggle to keep pace with rivals.
Alexandr Wang, Meta’s new chief AI officer, said in a post on X on Friday that the company planned to license Midjourney’s “aesthetic technology for our future models and products, bringing beauty to billions” in a “technical collaboration” between their research teams.
“To ensure Meta is able to deliver the best possible products for people it will require taking an all-of-the-above approach,” he added. “This means world-class talent, ambitious compute road map, and working with the best players across the industry.”
The tie-up will allow Meta to develop and integrate multimedia AI generation features into its apps, as chief executive Mark Zuckerberg has indicated that he expects AI-generated content to become more prominent on the platform.
The move comes as Zuckerberg pours billions of dollars into developing “superintelligence” that surpasses human intelligence. In recent months he has aggressively poached top AI researchers from competitors, doubled down on his investment in AI infrastructure and acquired AI voice company Play AI. Meta also took a stake in data labelling group Scale AI.
This week, Meta announced it was restructuring its AI group — recently renamed Meta Superintelligence Lab — into four distinct teams, the fourth overhaul in six months as it has struggled to solidify its organisational structure.
The Midjourney partnership marks a shift by Meta away from building all of its AI models and products in house, after its existing ones began to lag rivals.
In 2024 Meta rolled out an image generation tool called Imagine, which allows users to generate images from text prompts. Last October it shared a research paper on a movie generation model, Movie Gen, that will generate and edit videos based on text prompts. Meta promised to integrate it fully into Instagram in 2025.
However, the integration has yet to happen and industry insiders say the model already appears antiquated compared with Google’s Veo 3 and OpenAI’s Sora models, which have been released to consumers.
The social media company had also abandoned plans to publicly release its flagship Behemoth large language model, according to people familiar with the matter, focusing instead on building new models.
Meta had started using third-party models internally for tasks such as coding, according to multiple people familiar with the matter, as faith in its Llama models has waned.
San Francisco-based Midjourney, founded in 2021 by David Holz, has become one of the most popular image generation companies, despite its chief executive refusing venture capital and instead opting to self fund. In June, it released its video model V1, which allows users to generate a short video from an existing image.
Holz said in a post on X on Friday that “bringing sublime tools of creation and beauty to billions of people is squarely within our mission”, adding that Midjourney remained an “independent, community-backed research lab, with no investors”.
Read the full article here
