faceId selects a saved avatar; generation version and API version do not establish a streaming checkpoint ID.
Access
POST https://api.simli.ai/compose/token with x-simli-api-key and a compatible faceId, then use the SDK with P2P or LiveKit WebRTC transport. Account and face availability apply.
Regions
No model-specific processing-region guarantee established. Provider is based in Norway; its privacy policy lists international service providers, which does not establish Norway-only inference.
audio
SDK streaming input: PCM16 audio at 16 kHz. The general audio guide demonstrates mono conversion; compressed-file support on the static route is separate.Source for audio (opens in a new tab)
sessionDefaults
maxSessionLength defaults to 3,600 seconds and maxIdleTime to 300 seconds. These are configurable defaults; a hard maximum is not published in the schema.Source for sessionDefaults (opens in a new tab)
transports
P2P and LiveKit WebRTC. Current AsyncAPI names /compose/webrtc/peer_to_peer and /compose/webrtc/livekit. The migration guide uses an older p2p shorthand; prefer the current SDK.Source for transports (opens in a new tab)
connectionDeadline
P2P offer must arrive within 30 seconds; LiveKit join must complete within 30 seconds. These are connection deadlines, not inference latency guarantees.Source for connectionDeadline (opens in a new tab)
handleSilence defaults to true. The JavaScript guide recommends false when using listenToMediastreamTrack to avoid artifacts.Source for silence (opens in a new tab)
Standard published terms and privacy policy permit training and service-development use of user content and activity data. No enterprise opt-out or inference-specific deletion guarantee established here.Source for Training data policy (opens in a new tab)
monitoring
GET /history/sessions supports Unix start/end-time filters. GET /ratelimiter/sessions reports active sessions. These are account APIs, not additional models.Source for monitoring (opens in a new tab)
enrollment
POST /faces/trinity with an image and gsVersion=GSA_1.0. The returned face is used in later streaming requests; this is avatar enrollment, not a documented general fine-tuning API.Source for enrollment (opens in a new tab)
imageRequirements
Preprocessing accepts JPEG, PNG or WebP below 5 MB, at least 512×512. A front-facing head must occupy at least 15% of image height; the largest qualifying head is selected in group images.Source for imageRequirements (opens in a new tab)
preprocessing
POST /faces/trinity/preprocess returns a reframed PNG and TARGET-RATIO header. Pass the image and that value as target-ratio to enrollment to avoid repeating framing. Without it, enrollment frames the image first.Source for preprocessing (opens in a new tab)
preprocessingTiming
Provider estimates roughly 25 seconds for normal framing and 160 seconds with retry=true for higher quality. This is preprocessing time, not live rendering latency.Source for preprocessingTiming (opens in a new tab)
generationStatus
GET /faces/trinity/generation_status optionally filters face_id. Read generations for concurrent jobs; top-level fields describe only the latest. Completed and failed generation-status records remain listed for 24 hours. This is not content-retention policy.Source for generationStatus (opens in a new tab)
deletion
DELETE /faces/trinity/{id} removes a saved face and breaks agents depending on it. This is not proof that every related input, backup or training artifact is erased.Source for deletion (opens in a new tab)
freePlan
Free plan lists 50 minutes/month and 1 concurrent session. The homepage separately mentions $10 on signup; that is promotional credit, not a permanent inference tariff.Source for freePlan (opens in a new tab)
plans
Hobby: $10/month; 1,000 minutes/month; published $0.01/min; 2 concurrent sessions; 1 face generation. Pro: $49/month; 5,500 minutes/month; published $0.0095/min; 10 concurrent sessions; 5 face creation slots. Scale: $249/month; 27,500 minutes/month; published $0.009/min; 50 concurrent sessions; 30 face creation slots.Source for plans (opens in a new tab)
planConditions
Unlimited agents does not mean unlimited simultaneous sessions or usage minutes. Enterprise pricing is custom. Displayed per-minute rates are not explicitly labeled overage; effective bundle arithmetic may differ.Source for planConditions (opens in a new tab)
Prices retain their published currencies and billing units below.
Published billing units · original currency · observed 14 Oct 2026
Category
Usage
Rate
Unit
Trinity plan
Hobby displayed minute rate1
$0.01
minute
Trinity plan
Pro displayed minute rate2
$0.0095
minute
Trinity plan
Scale displayed minute rate3
$0.009
minute
Rate conditions
Published Hobby plan rate with $10/month subscription, 1000 included minutes/month and 2 concurrent sessions. Copied from the selected Trinity pricing panel, not derived by division. The card does not specify whether this is an overage tariff. No mapping to FastTalk/ARTalk or static-generation billing is assumed; connected speech/LLM services are separate.
Published Pro plan rate with $49/month subscription, 5500 included minutes/month and 10 concurrent sessions. Copied from the selected Trinity pricing panel, not derived by division. The card does not specify whether this is an overage tariff. No mapping to FastTalk/ARTalk or static-generation billing is assumed; connected speech/LLM services are separate.
Published Scale plan rate with $249/month subscription, 27500 included minutes/month and 50 concurrent sessions. Copied from the selected Trinity pricing panel, not derived by division. The card does not specify whether this is an overage tariff. No mapping to FastTalk/ARTalk or static-generation billing is assumed; connected speech/LLM services are separate.
Published privacy policy permits use of content and activity data for training/research. User content is retained as necessary; account data is deleted on request or within 12 months after account termination, while name/email have a separate 24-month period. Legal exceptions and connected providers apply; no default zero-retention promise.
faceId selects a saved avatar; generation version and API version do not establish a streaming checkpoint ID.
Access
POST https://api.simli.ai/static/audio with a compatible faceId, audioBase64 and audioFormat. Returns HLS and MP4 URLs. The generic request schema does not publish per-generation compatibility, duration or size ceilings.
Regions
No model-specific processing-region guarantee established. Provider is based in Norway; its privacy policy lists international service providers, which does not establish Norway-only inference.
audioSampleRate defaults to 16,000 Hz; audioChannelCount defaults to 1; videoStartingFrame defaults to 0. These defaults do not prove a fixed model sample rate or token context.Source for audioDefaults (opens in a new tab)
The old textToVideoStream route was removed. Generate TTS externally, then submit audio; no native Simli text-to-speech model is inferred.Source for speechInput (opens in a new tab)
Pricing terms
The public rate card does not separately specify static-file generation pricing or confirm that it matches live streaming. No standalone static tariff assigned.Source for Pricing terms (opens in a new tab)
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Published privacy policy permits use of content and activity data for training/research. User content is retained as necessary; account data is deleted on request or within 12 months after account termination, while name/email have a separate 24-month period. Legal exceptions and connected providers apply; no default zero-retention promise.
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
Simli
Not verified
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
How it works
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Minimum reusable input
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Lifetime & refresh
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
How to enable it
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Usage fields
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Billing rules
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Limits & data boundaries
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
Simli
Not verified
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
How it works
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Minimum reusable input
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Lifetime & refresh
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
How to enable it
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Usage fields
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Billing rules
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
Limits & data boundaries
Not collected
Persistent avatar faces, generation-status history and reusable session tokens are not a documented prompt-cache or discounted token-input contract.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
What the measurements say
No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
Simli publicly names Trinity-1 as its Gaussian avatar model. The provider retains proprietary service/algorithm rights; no public weight release or permissive weight license is established. The open SDK is client code, not the serving weights.
