Lemon Slice, a digital avatar startup founded in 2024, has launched a new diffusion-based model designed to add real-time video to AI agents and chatbots. The model, called Lemon Slice-2, generates digital avatars from a single image and enables interactive video conversations layered on top of a knowledge base.
The company said the 20-billion-parameter model can run on a single GPU and stream video at 20 frames per second. It is offered through an API and embeddable widget, allowing companies to integrate video avatars into applications with minimal setup. Avatars can be customized after creation, including changes to background, styling, and appearance.
Lemon Slice is positioning the technology for use cases such as customer support, education, language learning, e-commerce, and corporate training. The model supports both human and non-human avatars, with voice generation powered by ElevenLabs. The company said it has safeguards to prevent unauthorized face or voice cloning and uses large language models for content moderation.
To support development, Lemon Slice raised $10.5 million in seed funding led by Matrix Partners, with participation from Y Combinator and several technology executives. The startup plans to expand its engineering team and increase spending on AI compute.
