Why the demo matters more than the spec sheet
Most avatar SDK platforms publish impressive latency numbers. Under 1.5 seconds end-to-end. Thirty to sixty frames per second. Compatible with 99% of devices.
These numbers are meaningful — but they’re also easy to inflate with ideal test conditions. A demo running on your own device, on your own network, with your own ambient noise, tells you something the spec sheet cannot.
Before choosing an avatar SDK for a production integration, you should test three things in the live environment: how the avatar looks on a mid-range device, how it responds when you speak naturally (not in a clean, quiet room), and how quickly the conversation actually feels — not the documented latency, but the subjective experience of talking to it.
This guide walks you through how to test Spatius’s avatar SDK demo and what evaluation criteria actually matter for production use.
Try the live demo now
Spatius’s playground runs entirely in your mobile browser. No download, no account, no credit card.
Open that link on your phone. What you’re seeing is on-device rendering in action: the avatar computation is running on your device’s own hardware, not streaming from a cloud GPU. The server is sending approximately 10–20 KB/s of facial motion data; your device is doing the rest.
This is the architectural difference that separates Spatius from cloud-streamed competitors like HeyGen LiveAvatar and Anam. Those platforms send a video stream (1–2 Mbps); Spatius sends driving signals and renders locally. The practical result: the demo works on 4G, works on older phones, and costs a fraction of what cloud rendering costs at scale.
What to evaluate in any avatar SDK demo
When testing a live interactive avatar — whether Spatius’s playground or any competitor’s demo — look for these five things:
1. Perceived latency, not documented latency
Speak a natural sentence and measure the gap before the avatar begins responding. The human perception threshold for conversational lag is approximately 1.5 seconds; anything above that starts to feel unnatural. Spatius’s on-device rendering adds under 300 ms to the avatar layer — when combined with optimized ASR and TTS, total end-to-end latency is typically under 1.5 seconds.
Cloud-streamed avatar platforms add 400–800 ms on the rendering layer alone, before ASR and LLM processing.
2. Lip sync accuracy under natural speech
Speak quickly. Mumble slightly. Use a sentence with unusual phonemes. Cheap avatar rendering handles clean, slow speech tolerably well but breaks down on natural cadence. Watch whether the avatar’s mouth movements track your actual phonemes or run on a generic animation loop.
Spatius uses a proprietary lip sync algorithm that supports multiple languages — a hard requirement if your application serves non-English speakers.
3. Frame stability on a mid-range device
If you have access to a budget Android phone (under ~$200 USD), test there. Most cloud-rendered avatar demos look fine on any device because the rendering work is done server-side; the client only plays back a video. On-device rendering exposes device capability directly — and an SDK that can’t maintain 30fps on a mid-range device is not deployable to a general consumer audience.
Spatius supports adaptive quality levels across high, medium, and low resolution avatar models, maintaining stable 30–60fps down to budget hardware.
4. Layer separation and compositing
Does the avatar render with a transparent or separable background? For most real production scenarios — an avatar overlaid on a whiteboard, a PPT deck, a product catalog — the avatar needs to float over content as an independent layer. Cloud-based 2D avatar platforms typically require expensive multi-layer rendering add-ons. Spatius’s 3D architecture supports native layer separation out of the box.
5. SDK integration model
A live playground tests the rendering. But the real question for a production integration is: how cleanly does this fit into your existing AI stack? Spatius’s SDK is explicitly designed as an interface layer, not an AI pipeline. It accepts any ASR/LLM/TTS combination via WebSocket, LiveKit RTC, or Agora protocols — your AI logic stays entirely under your control.
For SDK documentation covering iOS, Android, and Web quickstarts: docs.spatius.ai
Spatius SDK: Core specs at a glance
| Metric | Spatius | Cloud-streamed alternative |
|---|---|---|
| Bandwidth per session | 10–20 KB/s | 1–2 Mbps |
| Additional rendering latency | <300 ms | 400–800 ms |
| Device compatibility | 99% Android/iOS | Requires strong network |
| Custom avatar creation | ~3 hours (3DGS) | Days to weeks |
| Free commercial avatar | High-fidelity avatar included | Varies |
| Platforms | iOS, Android, Web | Varies |
| LLM/TTS flexibility | Any stack | Often proprietary |
How to get API access
The playground is for evaluation. When you’re ready to integrate:
- Sign up at app.spatius.ai — create a developer account and get your API key. The free tier includes ~50 minutes of usage with no credit card required.
- Read the quickstart — iOS, Android, and Web SDK guides are at docs.spatius.ai
- Choose your transport — WebSocket for simpler integrations, LiveKit or Agora for production RTC architectures
When Spatius’s SDK is the right choice
- → Real-time, conversational avatar interaction (not pre-rendered video)
- → High-concurrency deployments where per-minute cloud GPU costs are prohibitive
- → Native mobile deployment (iOS/Android) at scale
- → A rendering layer that integrates with your own LLM and TTS stack
- → Custom, photorealistic avatar creation in hours rather than days
- → A no-code studio interface to produce training videos (see platforms like Synthesia)
- → A simple async avatar for one-off marketing content
- → A product that doesn't require developer integration
Want to explore further? Read our technical guides:
- How to Make a Hologram Out of an AI Avatar — a technical walkthrough for building floating, composited avatar displays
Ready to test? Open the Spatius playground now — no signup required. Try the playground , or ，或 Read the docs , or ，或 Get started free 。