Developers and businesses are increasingly deploying AI agents and chatbots within their apps, but so far they’ve been mostly limited to text. Digital avatar generation company Lemon Slice is working on adding a video layer to chats using a new pervasive model that can create digital avatars from a single image.
The model, called Lemon Slice-2, creates a digital avatar that runs on a knowledge base and performs any role an AI agent would need, including answering customer questions, helping with homework questions, and acting as a mental health support agent.
“In the early days of GenAI, my co-founders started experimenting with different video models, and it was clear to us that video was going to be interactive. The appealing thing about tools like ChatGPT was that it was interactive, and we want video to have that layer,” said co-founder Lina Colucci.
According to Lemon Slice, it is a 20 billion parameter model that runs on a single GPU and can live stream video at 20 frames per second. The company makes this model available through an API and embeddable widgets that businesses can integrate into their sites with a single line of code. After creating your avatar, you can change your character’s background, style, and appearance at any time.
In addition to human-like avatars, the company is also focused on being able to generate non-human characters to suit different needs. The startup uses Eleven Labs’ technology to generate the voices for these avatars.
Lemon Slice, founded in 2024 by Lina Colucci, Sidney Primus, and Andrew Weitz, is betting that it will differentiate itself from its competitors by using a unique general-purpose diffusion model (a type of generative model that learns to work backwards from noisy training data to generate new data) to create its avatars.
“The existing Avatar solutions that I have seen add negative value to the product,” Colucci says. “They’re creepy, they’re stiff. They look good for a few seconds, but as soon as you start interacting with them, they feel very creepy and don’t make you feel safe. The thing that’s holding Avatars back from really taking off is that there aren’t enough Avatars.”
tech crunch event
san francisco
|
October 13-15, 2026
To fund its efforts, the company announced Tuesday that it has raised $10.5 million in seed funding from Matrix Partners, Y Combinator, Dropbox CTO Arash Ferdowsi, Twitch CEO Emmett Shear, and The Chainblooming.
The company says it has guardrails in place to prevent unauthorized duplication of faces and voices and uses extensive language models for content moderation.
Lemon Slice declined to name the organizations using its technology, but said its model is used in use cases such as education, language learning, e-commerce and corporate training.
The startup faces stiff competition from video generation startups like D-ID, HeyGen, and Synthesia, as well as other digital avatar makers Genies, Soul Machine, Praktika, and AvatarOS.
Matrix partner Ilya Sukhar believes avatars will be useful in areas where video is prominent. For example, people prefer learning from YouTube rather than reading long texts. He pointed out that Lemon Slice’s technological strength and uniqueness will give it an advantage over other startups.
“We are a highly technical team with a track record of shipping machine learning products as well as demos and research.While many of the other players are customized to specific scenarios and industries, Lemon Slice takes a generalized (data and compute) “bitter lessons” scaling approach that has worked in other AI modalities. ” he said.
Y Combinator’s Jared Friedman believes that compared to other startups that focus on human-like avatars or game character-like avatars, Lemon Slice can generate all kinds of avatars by using a diffusion-style model.
“I think Lemon Slice is the only company with a fundamental ML approach that can finally overcome the uncanny valley and break the Avatar Turing test. “We’re training the same type of model, a video diffusion transformer. It’s a general-purpose model that does everything end-to-end, so there’s no upper limit to how good the results can be. It also works on both human and non-human faces, and you just need to add new faces.”
The startup currently has eight employees and plans to use the funding to hire engineering and go-to-market staff, as well as pay for computing fees to train its models.
