Computer vision applications are getting a huge boost from advances in smartphones combined with AI that helps fill in the gaps of what everyday devices cannot see. Using these tools, a startup out of Munich called Beyond Presence believes it holds the keys to what is coming next: Hyper-realistic avatars that look and sound exactly like their human counterparts and can be used in real-time, conversational situations.
The startup today said it raised its first outside funding, a pre-seed round of $3.1 million. German firm HV Capital is leading the round, with 10x Founders, Alba VC, and individuals from Meta, DeepMind, and Zalando also participating.
Beyond Presence hasn’t actually released its technology into the world just yet, but it is close, Awais Shafique, the startup’s CEO and co-founder, told TechCrunch.
Part of the new capital will be used to continue developing the company’s foundational models, he said, and some will fund the release of its public beta by the end of this month.
So far, some 300 companies are on the waitlist for the product, which will likely have early applications in areas like customer service and support, recruitment, sales, and e-learning. All of these are areas in which companies are looking to scale up interactions with users without having to hire and train more people.
So, how has a startup managed to raise a multimillion-dollar round even before it’s launched a product, and come this far as a resource-intensive AI company without any outside funding? The answer lies in the founders’ backgrounds.
Shafique previously co-founded Presize, a computer vision startup that could accurately take a user’s measurements by recording a clip of the user turning around once in front of a smartphone camera. Those measurements could then be used to help buy apparel online. The Munich-based startup had an outsized profile in its home country because the founders participated in the German TV version of “Dragons Den”/“Shark Tank” in 2020, where they raised a record-breaking €650,000.
Then in April 2022, it emerged that Meta had acquired Presize. The price and exact date of the transaction were never disclosed, but some strong signals point to a great exit.
SEC filings show that Meta spent between $774 million from January to March 2022, and $1.15 billion in the first six months of 2022, on acquisitions. Presize was the only deal publicly disclosed at that time. Wikipedia’s list of Meta acquisitions, meanwhile, has been edited with a much more modest estimate: $100 million.
Sources familiar with the matter told TechCrunch that a deal value in the lower nine figures was more accurate than 10 figures in this case.
Regardless of where Presize fell in that range, given that the startup had only raised “a couple of million,” in Shafique’s estimate, he and his co-founders exited very well.
“It’s lucrative enough that we don’t have to work,” he said in an interview.
(They’re all still working anyway. In addition to Shafique’s latest startup, Presize’s other co-founders, Leon Szeli and Tomislav Tomov, are still at Meta, working on its generative AI projects. They remain in contact with Shafique, though, as both are investing in Beyond Presence.)
At its heart, Presize was aiming to solve a thorny computer vision problem, and that is what Beyond Presence is aiming to do as well.
Admittedly, my initial thoughts about Beyond Presence’s pitch were skeptical.
Digital avatars are not exactly uncharted territory. Not only have they been around for decades, but they have been getting a boost of attention more recently — and maybe some hype — with the latest innovations in AI and processing, not to mention the craze for all things chatbot.
There are also no clear ideas of what will work in the realm of avatars as a long-term business. Some, like Hyper and Ready Player Me (backed respectively by Amazon and a16z), are drawing from gaming culture and the theory that avatars should not be realistic versions of the people using them. Those who want to hide their identities to preserve their privacy might opt for these approaches over those that replicate their exact likeness.
Others, like Synthesia, can create likenesses of people. But they require studio visits, and are not the basis of the main product, which uses modest devices like phone cameras to capture your expressions to power avatars that do not look like you.
And some others are trying to build tools that you can interact with using everyday devices — these include Veed and Deepbrain.
Beyond Presence is aiming for not only an avatar that can be created on-the-fly using your phone, but one that looks like you and can be used in real-time, unscripted conversational situations. Something like a video version of the text-based chatbots offered by OpenAI, Anthropic, Meta and many others.
The startup is not the only one in this space, of course. Zoom appears to be working on hyper-realistic custom avatars, and CommonGround is also developing something along similar lines.
For now, Beyond Presence is focusing on building something unique in the video component of its product.
Eleven Labs is powering the voice part of the experience, and OpenAI’s GPT is the generative AI model powering the content. Business customers can flexibly change which model they use.
“You can also leave everything up to us, [or use] any LLM that you would like, or any voice agent that you would like. And then we do the face on top of it,” he said. This feature addresses the fact that enterprise customers might already be working with particular LLM providers, for example, or do not want to share any proprietary data externally, he noted.
There’s an interesting detail about Beyond Presence’s early decisions around funding: It turns out that the company was accepted to participate in Y Combinator earlier this year, but opted out of the prestigious incubator.
Although YC definitely gives access to great networks, Shafique and his co-founder Felix Altenberger (CTO; and a former deep learning specialist at Presize) chose not to give up equity and autonomy to access that.
While YC partners might have had their own input on how to grow the company with sales, Shafique said he and Altenberger had a “horizontal” approach they wanted to pursue.
“We wanted to move more on the core technology and foundation model layer,” he said.
TechCrunch has an AI-focused newsletter! Sign up here to get it in your inbox every Wednesday.
Source: Techcrunch