Motion-data delivery
Speech audio is converted into a time-aligned driving stream. AvatarKit or another runtime applies that stream to locally loaded avatar assets.
Architecture comparison
Motion data vs video streaming for AI avatars
Motion data and video streaming deliver different artifacts. Motion data tells a client renderer how the avatar should move; video streaming delivers finished pixels. Choose based on which side should render, what the network can carry, and how much client control the product needs.
Verified Aug 3, 2026Decision-ready guidePrimary sources
Decision matrix
Compare the complete path.
Do not compare isolated numbers unless the definitions, inputs, environment, and included services match.
| Decision area | Motion-data delivery | Video streaming |
|---|---|---|
| Delivered artifact | Motion/control payload | Finished video frames |
| Visual customization | Client can combine avatar, UI, and background | Changes often require the rendered pipeline |
| Bandwidth | Published Spatius framing: 10–20 KB/s | Depends on codec, resolution, frame rate, and scene |
| Synchronization | Audio and motion timing at the client | Audio/video sync in the media stream |
| Client work | Asset loading and rendering | Media decoding and presentation |
| Failure mode | Missing motion, asset, or lifecycle state | Packet loss, jitter, encoder, decoder, or RTC state |
| Best fit | Embedded interactive applications | Centralized rendered experiences |
How it works
Two different operating models.
Architecture decides which team owns rendering, transport, recovery, and the surrounding AI product.
Video streaming
A cloud process renders the avatar, encodes the frames, and transports a media stream. The browser or app displays the decoded video.
Best fit
Choose for the system you can operate.
The best option is the one whose responsibilities match your product, client, network, and team.
Choose Motion-data delivery when…
- Need direct composition in the application UI
- Target networks make media costly
- Client runtime is controlled
- Want avatar delivery separate from the agent
Choose Video streaming when…
- Client only needs a media element
- No avatar runtime should ship to the client
- Central render output is preferred
- Existing RTC infrastructure handles media
Limitations and unknowns
A small payload does not remove the need for client performance testing. A video stream does not guarantee consistent visual quality under congestion. Measure the complete experience.
Build a controlled evaluation.
Use one workload and record both user experience and operational responsibility.
1. Evaluation stepCapture average and peak bytes in both directions.
2. Evaluation stepCompare time to first visible response.
3. Evaluation stepTest audio-motion or audio-video synchronization.
4. Evaluation stepDocument which system owns dropped-frame recovery.
- Define the user job and acceptable fallback.
- Use the same input, session duration, and client.
- Record latency, traffic, compute, errors, and recovery.
- Compare total operating cost, not only list price.
Evidence
Sources and freshness.
Last verified Aug 3, 2026. Recheck implementation details when SDKs or plan terms change.
Related decisions