Lemon Slice nabs $10.5M from YC and Matrix to build out its digital avatar tech

Developers and companies are increasingly deploying AI agents and chatbots within their apps, but so far these have mainly been limited to text. Digital avatar generation company Lemon slice is working to add a video layer to those chats with a new distribution model that can create digital avatars from a single image.

The model, called Lemon Slice-2, can create a digital avatar that works on top of a knowledge base and can play any role required of the AI ​​agent, such as answering customer questions, helping with homework questions or even working as a mental health support agent.

“In the early days of GenAI, my co-founders started playing with different video models, and it became clear to us that video would be interactive. The appeal of tools like ChatGPT was that they were interactive, and we want video to have that layer,” said co-founder Lina Colucci.

Lemon Slice says this is a 20 billion parameter model that can run on a single GPU to livestream videos at 20 frames per second. The company makes the model available through an API and an embeddable widget that companies can integrate into their sites with a single line of code. After an avatar is created, you can change a character’s background, style, and appearance at any time.

In addition to humanoid avatars, the company is also focusing on generating non-human characters to meet different needs. The startup uses ElevenLabs technology to generate the voices of these avatars.

Founded in 2024 by Lina Colucci, Sidney Primas, and Andrew Weitz, Lemon Slice is betting that using its own general-purpose diffusion model (a type of generative model that learns to work backwards on noisy training data to generate new data) to create avatars will set it apart from the competition.

“The existing avatar solutions I have seen so far add negative value to the product,” Colucci said. « They’re creepy and stiff. They look good for a few seconds, and then once you start interacting with them it feels very creepy and doesn’t put you at ease. What’s preventing avatars from really taking off is that they haven’t been good enough. »

To fund that effort, the company said Tuesday it has raised $10.5 million in seed funding from Matrix Partners, Y Combinator, Dropbox CTO Arash Ferdowsi, Twitch CEO Emmett Shear and The Chainsmokers.

The company says it has set guardrails to prevent unauthorized cloning of faces or voices, and uses large language models to moderate content.

Lemon Slice would not name which organizations are using the technology, but says the model is used for applications such as education, language learning, e-commerce and corporate training.

The startup faces stiff competition from video generation startups such as D-ID, HeyGen and Synthesia, as well as other digital avatar makers GeniusesSoul Machine, ExcerciseAnd AvatarOS.

Ilya Sukhar, partner at Matrix, thinks avatars will be useful in areas where videos are prominent. For example, people prefer to learn from YouTube rather than reading long blocks of text. He noted that Lemon Slice’s technical prowess and its own will give the company an edge over other startups.

“It’s a highly technical team with a track record of shipping ML products, not just demos and research. Many of the other players are tailor-made for certain scenarios or vertical markets, and Lemon Slice takes the overall “bitter lesson” scaling approach (to data and computing power) that has worked in other AI modalities,” he said.

Y Combinator’s Jared Friedman believes that by using a diffusion model, Lemon Slice can generate any kind of avatars, compared to some other startups that focus on human-like or game character-like avatars.

“Lemon Slice is, I believe, the only company taking the fundamental ML approach that can finally conquer the uncanny valley and break the avatar Turing test. They train the same type of model as Veo3 or Sora: a video diffusion transformer. Because it’s a general-purpose model that does everything end-to-end, it has no ceiling on how good it can get; the others top out below photorealistic. It also works for both human and non-human faces and only requires an image to add a new face,” he said.

The startup currently has eight employees and plans to use the money to hire technical and go-to-market staff, as well as pay for computing costs to train its models.