mcpField
T
tavus-replica

tavus-replica

github.com/tavus-engineeringCreative & media958 installsUpdated 39w ago
Repository

About tavus-replica

tavus-replica is a skill for creating and managing Tavus digital replicas used in video generation and real-time conversations. It covers stock replicas as well as custom replica training and lifecycle operations.

Key features

  1. 01Use ready-made stock replica IDs
  2. 02Prepare training and consent videos
  3. 03Create custom replicas through the API
  4. 04Check training status and list replicas
  5. 05Delete replicas through the API

Use cases

  1. 01Try video generation with a stock replica
  2. 02Train a replica from supplied video
  3. 03Track custom replica training progress
  4. 04Remove a replica that is no longer needed

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/tavus-engineering/tavus-skills --skill tavus-replica

After installing, type/tavus-replicain Claude Code to confirm it loaded.

Details

From the project README

Tavus Replica Skill

Create lifelike AI digital twins for video generation and real-time conversations.

Stock Replicas (Ready to Use)

No training needed. Use these IDs directly:

  • rfe12d8b9597 - Default stock replica
  • re8e740a42 - Nathan

Full list: https://docs.tavus.io/sections/replica/stock-replicas

Create Custom Replica

Training Video Requirements
  1. Duration: 1 min talking + 1 min silence (same video)
  2. Quality: 1080p minimum, stable camera at eye level
  3. Lighting: Diffuse, even lighting, no shadows on face
  4. Audio: Silent room, no background noise
  5. Framing: Face fills at least 25% of frame
  6. Expression: Start talking segment with a big smile
  7. Background: Simple, static, no moving objects/people
Consent Video

Required for personal replicas. Record yourself saying:

"I, [full name], consent to the creation of a digital replica of my likeness for use with Tavus technology."

Can be separate video or at start of training video.

API: Create Replica
curl -X POST https://tavusapi.com/v2/replicas \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_API_KEY" \
  -d '{
    "replica_name": "My Avatar",
    "train_video_url": "https://s3.../train.mp4",
    "consent_video_url": "https://s3.../consent.mp4",
    "callback_url": "https://your-webhook.com/tavus"
  }'

Response:

{
  "replica_id": "r783537ef5",
  "status": "training",
  "training_progress": "0/100"
}

Training takes 4-6 hours. You'll receive a webhook or poll status.

API: Check Training Status
curl https://tavusapi.com/v2/replicas/{replica_id} \
  -H "x-api-key: YOUR_API_KEY"

Status values: training, completed, error

API: List All Replicas
curl https://tavusapi.com/v2/replicas \
  -H "x-api-key: YOUR_API_KEY"
API: Delete Replica
curl -X DELETE https://tavusapi.com/v2/replicas/{replica_id} \
  -H "x-api-key: YOUR_API_KEY"

Best Practices

  • Use desktop recording apps (QuickTime, Windows Camera) not browser
  • Pre-signed S3 URLs must be valid for 24+ hours
  • Phoenix-3 model is default and recommended
  • Test with stock replicas before training custom ones

Comments

Markdown supported