Answers, without the wait.

Spatius bills Avatar sessions in Credits based on the time the session remains connected to Spatius servers. One Credit represents approximately 6 seconds of connection time, so 10 Credits are approximately equal to one Avatar minute. Billing starts when the connection is successfully established and stops when the session is disconnected or released.

A connected session continues to consume Credits even when the Avatar is temporarily silent. Creating a session token without establishing a connection does not consume session Credits. Connections shorter than 6 seconds are rounded up to 1 Credit. Custom Avatar Creation allowances are counted separately from session Credits.

Included Credits, concurrency limits, and session limits vary by plan; additional usage costs $0.01/min for all plans, regardless of monthly or annual billing. Please refer to the Pricing page. Enterprise plans start at USD 699 per month for 80 sessions and 40,000 minutes and can be bought self-serve up to 160 sessions; each session has a maximum price per month, and above 160 sessions please contact sales.

Billing is based on the total time the Avatar session remains connected, from successful connection to session end. Speaking, listening, and silence are all billed at the same rate.

Yes. Idle and listening time continue to consume Credits while the Avatar session is connected. Close the Avatar session to stop metering; muting the audio or hiding the Avatar does not stop billing.

Usage is rounded up to the next 6-second increment. One Credit equals 6 seconds, and 10 Credits equal 1 minute. Each session has a minimum charge of 1 Credit (6 seconds).

Starting a session requires a balance of at least 10 Credits (1 minute); this is a starting-balance requirement, not a 1-minute minimum charge.

Yes. You can close and restart the Avatar session within the same LiveKit call, while the Avatar continues rendering locally without an active Avatar session. Each restart creates a new Avatar session, with a new 6-second minimum charge and a fresh check for at least 1 minute of available balance.

We do not yet have a worked example for this pattern. We recommend starting without this optimization and considering it as usage grows; we will provide documentation and a demo to support it.

We do not publish a restart-time specification. Measure it in your own setup, and we can review the results with you.

A dropped connection is billed until the system detects the disconnection, so a short detection delay is included in the billed duration. Reconnecting creates a new session with its own 6-second minimum charge. If the old and new connections overlap, both are billed during the overlap.

In Spatius Studio, use session history, credit activity, and invoices to review usage and reconcile charges.

Additional usage costs $0.01/min for all plans, whether billed monthly or annually. This rate applies to usage beyond the plan's included allowance.

Refresh Studio and sign in again to confirm whether the plan status and balance have updated. If the Credits are still missing, provide the payment email, payment time, plan name, receipt or payment screenshot, and App ID. The support team will compare the Stripe payment record with the account entitlement, restore Credits when appropriate, and investigate the automatic fulfillment flow.

These errors usually mean the issuing bank or payment provider rejected the transaction, for example because of insufficient funds or because the card does not support this type of international or recurring online payment. Contact the issuing bank or try another eligible card. Spatius cannot remove a bank-side decline directly.

This usually indicates an unpaid Stripe invoice or an abnormal subscription state. Provide the subscription email, a screenshot of the page, and the latest payment record so the support team can verify the billing state and assist. Avoid repeatedly subscribing or retrying payment, as this may create duplicate orders.

Standard plans follow the concurrency limit shown on the applicable plan. Concurrency means the number of Avatar sessions that can remain active at the same time. Enterprise capacity can be configured according to business requirements. For capacity planning or a quotation, provide normal concurrency, peak concurrency, how long and how often peaks occur, and the estimated total monthly connection time.

The Free plan is intended for testing public Avatars and does not include custom Avatar Creation by default. Paid plans include creation allowances according to their plan terms. Users conducting a technical evaluation may contact support and describe their use case; trial creation allowances are assessed case by case, and no paid subscription will be activated without the user's confirmation.

Cancel any time in Studio. Monthly plans end at the end of the current month; annual plans run to their renewal date. Spatius does not refund payments already made, and unused Credits are not paid back. In exceptional cases, such as a fault on our side, support may issue an account credit. If the Cancel button is disabled, check the FAQ about outstanding invoices.

support@spatius.ai or the Support link in Studio. Include App ID, Avatar or connection ID, SDK version, integration mode and a recording for visual issues. Enterprise plans follow their agreed SLA.

Yes, per plan: maximum length of one session, concurrency (see the concurrency FAQ) and Credits per month. A session that hits its limit disconnects and can reconnect at once. Current numbers are on the Pricing page; Enterprise limits are set by contract.

No. Deleting a successfully created Avatar is not treated as a failed creation, so the Avatar Creation allowance is not automatically returned. If the creation task failed, the status is abnormal, or the result has a clear technical defect, keep the Avatar record and provide the Avatar ID and screenshots to support.

Do not delete it first, because the record may be needed to determine whether a retry, rebuild, or compensation is appropriate.

Keep the failed task and do not delete it immediately. Record the Avatar ID, failure time, source image, and a screenshot. Refresh the page to check whether the displayed status has synchronized. If the backend task succeeded while Studio still shows Failed, support may need to inspect the task status and, when necessary, rebind or rebuild the Avatar.

Use a clear, front-facing upper-body portrait: the entire head and hair visible, the head upright, and enough space between both shoulders and the image edges. In the shot, avoid strong body or face rotation, visible hands, and accessories such as stethoscopes or reflective jewelry. In the image, avoid dense patterns, heavy knitted textures, excessively curly hair, uneven lighting, and wide-angle or selfie-lens distortion; the subject should be sharp.

Spatius currently supports creating realistic, semi-realistic, stylized, and anime-style Avatars from a single image. For semi-realistic, stylized, and anime-style Avatars, the facial features should remain relatively human-like. Different visual styles and levels of image complexity may affect generation stability, so we recommend testing the target image before production use to verify the character’s appearance, lip movement, teeth, and edge quality.

Teeth and the mouth interior are difficult details to reconstruct, and artifacts may occasionally result from the source image, hair occlusion, or generation randomness. Provide the Avatar ID, a still screenshot, and a recording of the Avatar speaking. Support will evaluate whether a rebuild or regeneration is appropriate.

If repeated attempts remain abnormal, try a source image with more even lighting, a clearer mouth area, and less occlusion.

Creation currently takes about 3 to 5 hours, although the actual time depends on the processing queue, image quality, and task status. Refer to the progress shown in Studio. Once created, the Avatar's clothing, hairstyle, and overall appearance cannot be directly replaced; significant visual changes normally require a new source image and a new creation.

Post-processing parameters such as Expressiveness, Head, and Jaw can be adjusted in Studio.

When no session is connected or no audio is being supplied, the Avatar remains in an idle or breathing state. It may show subtle breathing, blinking, and natural micro-movements, but it does not continuously perform speaking mouth shapes. Lip-sync and speech-related expressions begin only after a connection is established and audio is supplied.

Expressions are currently generated primarily from the input audio. The public product does not yet provide emotion tags, gesture triggers, or RPC-level control for precise expression activation. Existing behavior is organized around states such as idle, listening, and speaking. If a project requires explicit emotional or gesture control, provide the required labels, trigger timing, and business scenario for product evaluation.

Emotion control is planned in the next generations of the product.

The standard product currently focuses on head-and-shoulders and upper-body real-time conversation. Full-body Avatars and programmable gestures are not standard capabilities. Some head, gaze, and subtle body movements are generated naturally from the state and audio, but a specific gesture cannot currently be selected through the API.

Contact Spatius for evaluation if full-body output or specific actions are mandatory. Upper-body gestures are planned in the next generations of the product.

If the Studio preview is normal, the issue is more likely in the integration layer: for example, the Canvas is mounted more than once, an old instance is not destroyed, multiple render layers overlap, or scaling is incorrect. Compare your implementation with the latest official demo, and check whether the page contains multiple Canvas elements or repeated initialization.

When submitting a ticket, include the SDK version, browser, device, recording, and a minimal reproduction.

Client SDK coverage includes Web, iOS, and Android, with a Flutter plugin also available. Web rendering relies on the client GPU and WebGPU. Use a current version of Chrome or Edge where possible, and test on the intended phones, computers, WebViews, or embedded devices. Browser compatibility and minimum hardware requirements may change with SDK releases, so refer to the current official documentation.

Spatius renders the Avatar with a transparent background on a Canvas. The customer application should provide the background: for example, an image, video, or page background placed underneath the Avatar Canvas. Ensure that the background layer sits behind the Canvas, matches the container size, and does not cover the character.

Do not treat the background as a video layer produced by the SDK.

Use the positioning and camera controls provided in Studio whenever possible, and validate the result on the target device. If virtual-camera parameters are adjusted in code, use only interfaces supported by the current SDK documentation. Overwriting the original camera data directly may cause the Avatar to shrink, shift, or disappear after reconnecting.

No. Spatius does not render video in the cloud and deliver it as a conventional video stream. It sends audio and motion data, which AvatarKit renders in real time using the client GPU. The motion track therefore cannot be decoded by a standard <video> element; the Spatius client SDK is required for decoding and rendering.

Spatius currently supports four primary integration modes: Direct Mode, Backend Mode, LiveKit Agents, and Agora Conversational AI. All four drive the Avatar with audio and rely on the client SDK for final rendering. Their main difference is how audio and motion data enter and travel through the system.

Direct Mode

The client connects directly to Spatius and sends audio. It requires the least integration work, but depends more heavily on client network conditions and typically has higher latency than the RTC or Backend paths.

Best for: Fast validation and Web, iOS, or Android applications that prefer a simple client-direct architecture.

Backend Mode

The customer's backend or Agent Server manages the session and sends audio to Spatius. It offers lower latency and centralized control over authentication, audio streaming, interruption, and business logic.

Best for: Production applications with an existing backend, WebSocket pipeline, voice agent, or a requirement for full transport control.

LiveKit Agents

The official LiveKit Agents plugin publishes audio and motion data into a LiveKit Room. It provides low latency and reuses existing LiveKit Agent orchestration. AvatarKit is still required on the client; this is not a conventional video stream.

Best for: Projects already using LiveKit Agents. Check the current SDK documentation for platform-specific support.

Agora Conversational AI

The Agora Conversational AI Provider and Agora Channel carry synchronized audio and motion data, reusing Agora's RTC network and synchronization capabilities.

Best for: Projects already using Agora RTC or Agora Conversational AI.

Direct Mode is usually the fastest way to validate an integration. If the project already has a backend, LiveKit, or Agora infrastructure, use the mode that best matches the existing architecture. See the Integration Overview for current support and implementation details.

Yes. Spatius consumes the audio stream produced by the customer's system and uses it to generate motion data and Avatar behavior. ASR, dialogue logic, LLM, TTS, voice models, and the knowledge base can remain under the customer's control. Language support therefore depends mainly on the selected TTS and audio quality, rather than on a dialogue model built into Spatius.

This is often not a GPU frame-rate problem. Instead, the supply of animation frames has gaps: audio continues to play locally in the application, while motion data arrives in batches and the image pauses briefly between batches. Send each audio chunk as soon as it is generated instead of accumulating a full sentence or artificially throttling transmission to playback speed.

Where supported, try setting frameStarvationMode to strictSync so the video-frame schedule follows the audio more strictly.

Not necessarily. A natural mouth opening, breath, or transition animation may begin before clearly audible speech. If a persistent and obvious offset remains, provide the connection ID, audio format, sample rate, chunking strategy, and a recording. Check whether too much audio is sent in advance, whether audio playback is independently scheduled by the customer application, and whether the end-of-audio signal is sent correctly.

First confirm that audio actually enters the Spatius driving pipeline instead of playing only on the local device. Then check the RTC provider, App ID, region, and channel permissions. For example, an Agora App ID that cannot access the target regional RTC service may load the Avatar but fail to receive driving data.

Provide the App ID, connection ID, SDK version, integration mode, and browser console logs when contacting support.

Real-time recording consumes resources for both client rendering and video encoding. First reduce the rendering resolution or quality through RenderQuality, and perform a load test on the target device. For offline video export or broadcast workflows, contact Spatius with the target resolution, frame rate, audio tracks, background requirements, concurrent workload, and output protocol so the appropriate supported workflow can be evaluated.

About 700-1200 ms end to end on Backend Mode, LiveKit Agents or Agora Conversational AI; Direct Mode is slower and depends on the client network. Your own ASR, LLM and TTS add to that. If you measure much more, send the connection ID, integration mode, region and a recording.

Under the Spatius Terms of Service, customers retain ownership of the photos, audio, and other Customer Content they submit, as well as the Outputs generated through the Platform. Customers must also have explicit authorization from every person depicted for the intended use. A custom Avatar is associated with the account that created it and is not automatically made public or available to other developers.

Other developers may use a corresponding public copy only after the customer explicitly agrees to publication. Ownership of an Output does not mean that standard plans include delivery of downloadable raw 3DGS models, training intermediates, or underlying runtime assets; normal access is through Spatius Studio, SDKs, and APIs.

Export, exclusivity, migration, or source-asset delivery must be governed by a separate written agreement. Public Avatars are platform resources licensed for use, not assets owned by the customer. See the Terms of Service for details.

The Spatius cloud receives audio and generates motion data, while the client plays audio and renders the Avatar. The processing region depends on the integration mode and selected region. The standard product is currently cloud-based. Confirm the available regions for your integration before deployment.

Contact Spatius for private deployment or country-specific hosting requirements. The same applies to data residency, cross-border transfer, financial, educational, or other compliance requirements.

US West (Silicon Valley), Japan (Tokyo) and Mainland China (Beijing) today; EU and Singapore regions are planned. Choose the region closest to your users. If your region is not covered yet, contact support with the country and expected volume.

Spatius processes audio only to generate motion data and never trains on customer content. Our GDPR programme (the Article 28 data processing agreement and supporting documentation) is in progress; a DPA will be available on request once finalised. Spatius does not hold a SOC 2 or ISO 27001 certification at this time.