[ Architecture · Evaluation ]
Evaluate Atlas
on the output.
Atlas is the rendering layer for realtime and async avatar video. Review the output showcase, integration surface, and operating model before deciding whether it fits your product.
Showcase outputs use a single reference image without per-avatar training or fine-tuning.
01
One-shot
Avatar setup
A single reference image; no per-avatar training step.
02
$7/hr
Rendering rate
Realtime and async output, prorated by duration.
03
WebRTC
Realtime delivery
Audio in, lip-synced video out.
04
BYO stack
Conversation layer
Keep your STT, LLM, TTS, tools, and application logic.
One reference image, two delivery modes
Use the same avatar-rendering API for a live WebRTC session or a finished asynchronous video.
01
Realtime passthrough
Create a session, publish audio from your application, and receive a synchronized video stream over WebRTC.
- Your application owns STT, LLM, TTS, and tools
- No fixed session-length cap while the session remains active
- Multi-viewer tokens are available for shared playback
02
Async generation
Submit a reference image and audio, poll the job, and download the completed video from a time-limited result URL.
- REST job workflow with status and result endpoints
- Webhook delivery with signed payloads
- Output-duration billing at the same rendering rate
[ Product boundary ]
You own the brain. Atlas renders the face.
Your application
- Speech recognition and text-to-speech
- LLM, memory, tools, and moderation
- Prompts, customer records, and business logic
Atlas
- Reference-image avatar rendering
- Lip synchronization from supplied audio
- Realtime WebRTC or asynchronous video delivery
Test it with your workload.
Visual quality depends on the reference image, audio, and target experience. Use the public outputs as a starting point, then test your own inputs through the API.