Tavus

Tavus specializes in personalized video generation at scale: train a "replica" (AI clone of voice and appearance) from a short training video, then generate thousands of unique videos where that replica delivers a customized script per recipient. For agents automating video outreach: create a video generation job with a replica ID, script, and custom variables — Tavus renders asynchronously and notifies via webhook when ready. Primary use cases: personalized sales videos, automated customer onboarding, dynamic marketing content. Tavus CVI for real-time AI avatar conversations. Confidence is docs-derived.

REST API at tavusapi.com/v2. Resources: replicas, videos, conversations (CVI). GET /v2/replicas lists available replicas. POST /v2/videos creates a video generation job ({replica_id, script, video_name, callback_url, custom_data}). GET /v2/videos/{video_id} polls status (queued → generating → ready/failed). Status: queued → generating → ready (download_url populated). DELETE /v2/videos/{video_id} removes a generated video. POST /v2/conversations creates a real-time CVI session.

API key auth: x-tavus-api-key header. Keys from Tavus Dashboard → Settings → API Keys. Account-scoped. HTTPS enforced. No OAuth2. Video download URLs are presigned (time-limited access). Replica training requires uploading a 2–5 minute consent video; Tavus applies identity verification checks before activating a replica.

Standard HTTP status codes with JSON error bodies. 401 for invalid API key, 400 for missing required fields. Video generation is async (typically 1–5 minutes). Poll GET /v2/videos/{id} every 10–30 seconds. Callback URL receives JSON payload on ready or failed status. Failed videos include error_message field. Rate limits: plan-dependent; generating many videos simultaneously may queue. Tavus uptime at status.tavus.io. Video quality depends on replica training data quality.

docs.tavus.io provides API reference, replica creation guide, video generation walkthrough, and CVI documentation. Getting started: create Tavus account, train a replica (requires training video upload), first video generation in sandbox in under 30 minutes. Direct REST API usage (no official SDK beyond API reference). Webhook integration guide included. Documentation quality is solid for a specialized AI video platform.