TL;DR
Lemon Slice closed a $10.5 million seed round led by Matrix and Y Combinator backers to develop video-capable digital avatars. The startup unveiled Lemon Slice-2, a 20-billion-parameter diffusion model that can generate avatars from a single image and stream them via an API and embeddable widget.
What happened
Lemon Slice, founded in 2024 by Lina Colucci, Sidney Primas and Andrew Weitz, announced a $10.5 million seed round from investors including Matrix Partners, Y Combinator, Dropbox CTO Arash Ferdowsi, Twitch CEO Emmett Shear and The Chainsmokers. The company is building a video layer for AI agents with a new model called Lemon Slice-2, a diffusion-based system that it says contains roughly 20 billion parameters and can run on a single GPU to livestream at about 20 frames per second. The model can create an avatar from a single still image, supports live changes to background and styling, and is offered through an API and a one-line-embed widget. Lemon Slice uses ElevenLabs technology for avatar voices, employs large language models for moderation, and reports guardrails against unauthorized face or voice cloning. The startup currently has eight employees and plans to use the funding for engineering hires, go-to-market staff and compute costs.
Why it matters
- Adding a video layer to chatbots could make AI agents more engaging where visual content is important, such as education and e-commerce.
- Single-image avatar creation lowers friction for adoption, enabling faster onboarding compared with multi-shot capture pipelines.
- A general-purpose diffusion approach may scale improvement across many character types, potentially reducing the 'uncanny valley' in interactive video agents.
- Built-in safety controls and LLM-based moderation address key ethical and misuse concerns, though implementation details remain sparse.
Key facts
- $10.5 million seed round announced, led by Matrix and Y Combinator investors
- Investors also include Arash Ferdowsi (Dropbox CTO), Emmett Shear (Twitch CEO) and The Chainsmokers
- Model name: Lemon Slice-2 — described as a diffusion model with ~20 billion parameters
- Company claims the model can stream video at ~20 frames per second on a single GPU
- Avatars can be generated from a single image and edited later (background, styling, appearance)
- Available via an API and an embeddable widget designed for single-line integration
- Voice synthesis for avatars leverages ElevenLabs technology
- Lemon Slice has eight employees and will use funding for hires and compute
- Reported use cases include education, language learning, e-commerce and corporate training
- Company says it uses guardrails to prevent unauthorized face/voice cloning and employs LLMs for content moderation
What to watch next
- Availability and uptake of the API and embeddable widget among enterprise customers (confirmed in the source)
- Which organizations publicly deploy Lemon Slice in education, e-commerce or training (not confirmed in the source)
- Progress toward overcoming the uncanny valley and reaching photorealism at scale (not confirmed in the source)
- Regulatory and platform responses to face and voice cloning guardrails and content-moderation practices (not confirmed in the source)
Quick glossary
- Diffusion model: A type of generative model that learns to produce new data by reversing a noising process applied during training; used for images and video generation.
- Parameter: A numeric value in a machine learning model that is learned from data; models with more parameters typically have greater capacity.
- Avatar: A digital character or representation, often used to personify an AI agent or user in visual interfaces.
- API (Application Programming Interface): A set of rules and endpoints that lets developers integrate external services and functionality into their own applications.
- Uncanny valley: A phenomenon where near-realistic human representations provoke discomfort because they fall short of full realism.
Reader FAQ
How does Lemon Slice create avatars?
The company uses a diffusion-based model called Lemon Slice-2 that it says can generate an avatar from a single image and stream video at about 20 fps on a single GPU.
Who invested in the seed round?
Investors named include Matrix Partners, Y Combinator, Arash Ferdowsi, Emmett Shear and The Chainsmokers.
Can the system clone someone’s face or voice without permission?
Lemon Slice says it has guardrails to prevent unauthorized face or voice cloning and uses large language models for moderation.
Which companies are using Lemon Slice’s technology?
Not confirmed in the source.
What are the pricing and commercial terms?
Not confirmed in the source.
Developers and companies are increasingly deploying AI agents and chatbots within their apps, but so far they’ve mostly been restricted to text. Digital avatar generation company Lemon Slice is working…
Sources
- Lemon Slice nabs $10.5M from YC and Matrix to build out its digital avatar tech
- 💻 Lemon Slice Raises $10.5M to Build Digital Avatar Tech
- Lemon Slice nabs $10.5M for avatar tech. Summary
- Breaking News: Lemon Slice nabs $10.5M for avatar tech.
Related posts
- Marissa Mayer’s Dazzle raises $8M seed led by Forerunner’s Kirsten Green
- Founder of Atmos raises $1.65M for Drafted, an AI custom-home design tool
- Amazon’s Alexa+ adds Angi, Expedia, Square and Yelp integrations