LemonSlice is an AI-avatar service for building interactive avatars that listen, talk, and respond to a user in real time within a video call. Unlike a static video avatar that plays back pre-recorded clips, it generates each frame live, so lip movement, facial expression, and body language follow the conversation as it happens.
An avatar is created from a single uploaded photo, without training, fine-tuning, or a character rig, and any image with a face, including non-human characters, can be animated. A language model, voice provider, and knowledge base are connected through LemonSlice's API, and the resulting avatar is deployed to a website or product as a widget, a hosted pipeline, or a self-managed pipeline.
The service is built on a Character World Model, trained to animate a character's face and movements in real time, and the project states a measured p99 response latency of 471 milliseconds, with support for over 1,000 concurrent sessions and calls lasting up to 24 hours.
Features
- Photo-to-avatar: an interactive avatar is generated from a single uploaded image with no training or rigging step
- Model and voice choice: connects to a language model, voice provider, and knowledge base through the API
- Deployment options: ships as an embeddable widget, a hosted pipeline, or a self-managed pipeline
- Low-latency response: measured p99 latency of 471 milliseconds
- Concurrency and session length: supports over 1,000 concurrent sessions and calls of up to 24 hours
