What Is a Replica?
A Replica is a hyper-realistic AI-generated video avatar created using Phoenix, Tavus’s rendering model. Phoenix is built on a Gaussian‑diffusion architecture. It delivers high-fidelity rendering with expressive output.Emotional expression controls in conversational video apply only to Phoenix-4; see Emotion Control with Phoenix-4.
voice_name using the API (see Replica Training); that path is faster to set up but lower fidelity than video. Prebuilt presenters are available as stock replicas.
Key Features
Realistic Face Cloning
Replicates a person’s look, expressions, and speaking style.
Multilingual Support
Enables natural conversations in up to 30 languages with accent preservation.
Reusable
Trained Replicas can be reused without new training video or images.
Replica Types
| Type | Description | Requirements |
|---|---|---|
| Personal | A digital human modeled after a real person’s facial appearance and voice. | Verbal consent (video training) or image rights (image training) — Replica Training |
| Non-Human | A digital human modeled after an AI-generated character. | No consent required |
| Stock | A prebuilt, professional digital presenter optimized for natural conversation. | No consent required |
Getting Started
You can create a personal or non-human replica using the Developer Portal or by following the steps in the Quickstart Guide. Training data can be a video, an image (with a stock voice), or—for non-human use cases—video from an AI-generated character; see Replica Training for API fields and requirements.Creating a Personal Replica is only available on the Starter, Growth, and Enterprise plans.