Lemon Slice secures $10.5M from YC and Matrix to build its own digital avatar technology

Developers and companies are increasingly deploying AI agents and chatbots within their apps, but so far they have been mostly limited to text messaging. Digital avatar generation company Lemon slice It’s adding a video layer to those chats with a new publishing model that can create digital avatars from a single photo.

The model, called Lemon Slice-2, can create a digital avatar that runs on top of a knowledge base to play any role required of an AI agent, such as answering customer inquiries, helping with homework questions, or even acting as a mental health support agent.

“In the early days of GenAI, the co-founders started playing with different video models, and it became clear to us that video was going to be interactive. The compelling part about tools like ChatGPT was that they were interactive, and we wanted 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 stream live video at 20 frames per second. The company is making the model available through an application programming interface (API) and an embeddable tool that businesses can integrate into their sites with a single line of code. After creating your avatar, you can change the background, design, and appearance of your character at any time.

Besides human-like avatars, the company is also focusing on the ability to create non-human characters to suit different needs. The startup uses ElevenLabs technology to create the voices of these avatars.

Lemon Slice, founded by Lina Colucci, Sydney Primas and Andrew Weitz in 2024, is betting that using its general-purpose diffusion model (a type of generative model that learns to work backwards from noisy training data to generate new ones) to make avatars will set it apart from competitors.

“The current avatar solutions I’ve seen so far add passive value to the product,” Colucci said. “It’s scary, and it’s cruel. It looks good for a few seconds, and once you start interacting with it, it feels very weird, and it doesn’t feel good. The thing that kept the avatars from really taking off was that they weren’t good enough.”

To fund the 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 guardrails in place to prevent unauthorized facial or voice cloning, and that it uses large language models to moderate content.

Lemon Slice did not name the organizations using its technology, but said the model is being applied in use cases such as education, language learning, e-commerce and corporate training.

The startup faces stiff competition from video production startups like D-ID, HeyGen, and Synthesia, as well as other digital avatar makers. Genessoul machine, He practicesand AvatarOS.

Ilya Sukhar, partner at Matrix, believes avatars will be useful in areas where videos stand out. For example, people like to learn from YouTube rather than reading long blocks of text. He noted that Lemon Slice’s technical prowess and special skills will give it an advantage over other startups.

“It’s a deep technical team with a proven track record of shipping ML products, not just demos and research. Many other players are tailored to specific scenarios or verticals, and Lemon Slice takes the overall “Bitter lesson” “The scaling approach (to data and computing) that has worked for other AI methods,” he said.

Y-Combinator’s Jared Friedman believes that using the Diffusion Pattern model allows Lemon Slice to create any type of avatar compared to some other startups that focus on human-like avatars or game character avatars.

“I think Lemon Slice is the only company with a fundamental approach to machine learning that can finally overcome the uncanny valley and break the Turing test for avatars. They train the same kind of models as Veo3 or Sora: the video post transformer. Since it’s a general-purpose model that does the whole thing from start to finish, it has no ceiling on how good it can be; the other models are less realistic. They also work with both human and non-human faces and only require an image to depict them. Add a new face,” he said.

The startup currently has eight employees, and plans to use the funds to hire engineers and market-specific staff, along with paying computing bills to train its models.