Compare real-time avatar APIs by published price, speed, and creative control.
Editorial order · Published claims · Definitions & sources
Swipe to compare more columns →
| # Editorial position | Model | Published rate ⓘ | Published latency ⓘ | Animation | Moderation | Evidence |
|---|---|---|---|---|---|---|
| 1 |  Sequencer AnimaFeaturedAnima Fast / Live |
$5.90/hrFast complete avatar | <1.8 sPublished response-latency claim | Custom signature motions | ConfigurableUncensored mode* | Source ↗ |
| 2 |  TavusPhoenix 4.5 / CVI |
$21.00/hrBuilder overage | 134 msAudio-to-video rendering | Expressive motion | Not verified | Source ↗ |
| 3 |  AnamCARA-4 |
$9.60/hrStarter overage | 180 msMedian server-side latency | Full-image rendering | Not verified | Source ↗ |
| 4 |  HeyGenLiveAvatar |
$11.40/hrEssential Full overage | <300 msFirst frames, vendor claim | Source-recording gestures | Not verified | Source ↗ |
| 5 |  Beyond PresenceGenesis 2.0 / A2V |
€21.00/hrStarter managed overage | ~100 msSpeech-to-video model | Natural expressions | Not verified | Source ↗ |
| 6 |  D-IDV4 Expressive Visual Agents |
Not publishedPlan dependent | <120 msCore model latency | Sentiment control | Not verified | Source ↗ |
| 7 |  RunwayCharacters / GWM-1 |
$12.00/hrSession meter + $0.02/start | Not listedComparable response timing not published | Prompted characters | Not verified | Source ↗ |
| 8 |  LemonSliceLemonSlice-2.1 / CWM-1 |
$18.60/hrStarter hosted overage | 471 msPublished p99 claim; scope varies | Actions & emotions (Ultra) | Not verified | Source ↗ |
| 9 |  SimliEmotive Faces / Trinity |
Not publishedNot published | <300 msPlatform speech-to-video estimate | Gaussian / emotive faces | Not verified | Source ↗ |
| · |  MuseTalkMuseTalk 1.5 |
Your computeYour GPU costs | 30+ FPSThroughput on NVIDIA Tesla V100 | Lip sync | App controlled | Source ↗ |
| · |  Quark LiveAvatarLiveAvatar 1.1 |
Your computeYour GPU costs | 45 FPSMulti-card H800, 4-step sampling | Streaming diffusion | App controlled | Source ↗ |
| · |  NVIDIAAudio2Face-3D |
Your computeYour compute costs | Not listedNo comparable timing in cited repository | 3D facial animation | App controlled | Source ↗ |
| · |  HedraRetiredRealtime Avatar (retired) |
UnavailableService retired | UnavailableLive service has been disabled | Historical streaming avatar | Service retired | Source ↗ |
Try a provider name or reset the filters.
Published entry usage rates, with the billing basis preserved.
Selected pay-as-you-go and entry-plan overage quotes.
Anima has the lowest entry usage quote in this selected USD set. Monthly fees, included minutes, session starts, model tiers, and enterprise discounts can change total cost.
Separate rendering speed from the time a user waits for a reply.
Beyond ~100 ms speech-to-video. D-ID <120 ms core model. Tavus 134 ms audio-to-video.
Anam 180 ms median server-side. HeyGen <300 ms first frames. Simli <300 ms speech-to-video.
Anima <1.8 s response claim. Anam <1 s median conversation. D-ID <500 ms conversation. LemonSlice 471 ms p99 claim.
Different clocks, networks, versions, and percentiles prevent a controlled fastest-model ranking. No comparable independent latency test has been run here.
| Provider | LSE-D ↓ | LSE-C ↑ |
|---|---|---|
| D-ID | 7.16 | 8.67 |
| Anam | 7.84 | 8.20 |
| HeyGen | 7.93 | 7.29 |
| Tavus | 8.42 | 6.33 |
D-ID describes five avatars per provider. Competitor model versions and equivalence to current releases are not established here. Anima was not included.
Original benchmark ↗Expand a model for specifications, limits, and primary evidence.
Sequencer AnimaAnima Fast / LiveCustom animation & moderationAnima Live provides conversational avatars with custom signature animations and configurable moderation. The Fast tier lists $5.90/hour for complete conversations and $5.31/hour for video-only.
Uncensored mode and custom filters are advertised; core safety limits remain. The <1.8 s figure is a vendor claim, with no controlled cross-provider test. It does not establish a fastest-model result.
TavusPhoenix 4.5 / CVIComplete conversation stackPhoenix 4.5 combines expressive rendering with the CVI conversation stack, perception, and turn-taking.
134 ms is rendering latency, not a full conversational turn. Starter has no overage; the hourly comparison uses Builder at $0.35 per extra minute.
AnamCARA-4Flexible conversational APICARA-4 supports image-based avatars, a modular JavaScript API, and your own language model.
$12/month Starter includes 50 minutes, then $0.16/min. Five-minute Starter sessions; published server and conversational latency measure different stages.
HeyGenLiveAvatarAvatar-only integrationLiveAvatar offers Full conversations or Avatar Only / Lite integration with your voice-agent pipeline.
Custom gestures follow source footage. Essential includes 550 Full or 1,100 Lite minutes, before overage; 1080p depends on avatar and plan.
Beyond PresenceGenesis 2.0 / A2VEnterprise deploymentsGenesis 2.0 offers speech-to-video rendering, managed agents, and enterprise private deployment options.
Starter includes 280 video-only or 140 managed minutes. Overage is €0.175/min video and €0.35/min managed. Rates remain in EUR.
D-IDV4 Expressive Visual AgentsPublished lip-sync resultsV4 Expressive Visual Agents supports emotion-aware delivery, optional vision, and interactive UI.
The SyncNet comparison is D-ID-run. Current competitors may use newer models than the tested snapshots. Pricing and output limits depend on access.
RunwayCharacters / GWM-1Interactive charactersGWM-1 powers custom conversational characters with session prompts, tool calling, and camera or screen inputs.
$12 is 60 minutes of metered usage, excluding $0.02 session starts and taxes. External voice-agent components may add costs.
LemonSliceLemonSlice-2.1 / CWM-1Character styles & actionsAnimate human and stylized characters from images. Ultra adds CWM-1 action and emotion controls; standard plans include LemonSlice-2.1.
The Starter hosted rate is $0.31/min, before its subscription. CWM-1 requires an eligible tier. Higher-volume and other-model prices can be lower.
SimliEmotive Faces / TrinityFocused speech-to-video APISimli adds an audio-driven face to an existing voice agent, with newer Gaussian Emotive Faces options.
The pricing modal lists $0.01/min for Emotive Faces Hobby and $0.05/min for Trinity Hobby. A comparable full voice-stack tariff is not established. The <300 ms homepage claim does not identify a model version.
MuseTalkMuseTalk 1.5 · Tencent Music EntertainmentSelf-hosted lip syncA practical open-source starting point for audio-driven lip synchronization. The project reports real-time throughput on a V100 after avatar preparation.
It edits the face region. Throughput depends on GPU, preprocessing, and output handling; it is not a hosted conversational API.
Quark LiveAvatarLiveAvatar 1.1 · AlibabaOpen streaming diffusionAn open research system for long-form audio-driven avatar video. Its 14B diffusion model reports 45 FPS on a multi-GPU H800 setup.
Single-GPU availability does not imply the reported multi-GPU real-time speed. This Quark project is separate from HeyGen LiveAvatar.
NVIDIAAudio2Face-3D3D character animationConsider Audio2Face-3D for game characters and stylized digital humans. It turns audio into facial animation that can be used in a 3D character pipeline.
This is a different output category from photorealistic video APIs. You supply the character, rendering, and conversation infrastructure.
HedraRealtime Avatar (retired)Retired live serviceHedra previously offered Realtime Avatars for interactive conversations through LiveKit. LiveKit now marks the service as disabled, and its integration no longer works.
Listed for historical coverage, outside the active API ranking. Current Hedra Avatar and Character 3 generate recorded talking videos; they are separate from the retired realtime service.
Anima is our featured editorial recommendation for custom animations, moderation control, and published entry usage pricing. The remaining order is a qualitative shortlist by integration fit and documented capabilities. It is separate from benchmark quality scores or a global cheapest / fastest claim.
All linked sources were rechecked October 13, 2026. Rates keep their currency, mode, and subscription basis. Retired products appear in the Retired and All views and have no active rate or rank. Unknown values remain unknown. A feature marked “Not verified” has no verified control claim in the sources checked; it does not mean the provider lacks the feature. Vendor results have not been independently reproduced.
Use identical audio and scripts, named model versions, the same region and network, and at least 30 turns. Report cold/warm starts, median and p95 complete responses, rendering latency, lip sync, interruptions, sustained FPS, reliability, and total cost. Download the blank benchmark template ↗
Anima Fast is our featured editorial pick for custom signature animations, configurable moderation, and its published entry usage rate. Tavus, Anam, HeyGen, Beyond Presence, D-ID, Runway, LemonSlice, and Simli suit different workflows. The order is a recommendation, with no independently measured quality score.
Anima Fast lists $5.90/hour for complete conversations and $5.31/hour for video-only. It is the lowest entry-tier full-conversation usage quote among the six USD examples shown in our pricing chart. Subscription bundles, enterprise discounts, other model tiers, and self-hosted compute can change the cheapest option for your workload.
Published numbers use different clocks: rendering, server processing, first frames, or a complete response. Anima lists under 1.8 seconds response latency. Tavus lists 134 ms rendering, Anam 180 ms median server latency, and Beyond Presence roughly 100 ms speech-to-video. These claims do not establish an apples-to-apples fastest-model winner.
Anima advertises uncensored mode and configurable moderation. Its published page also states that core safety limits remain. Review provider policies, model behavior, and deployment settings for the exact controls available to your use case.
This page compiles vendor documentation and project results. The SyncNet table was published by D-ID, and we have not independently reproduced it. A controlled evaluation needs the same inputs, model versions, hardware, network, and end-to-end timing for every provider.
MuseTalk and Quark LiveAvatar provide self-hostable code and weights. NVIDIA Audio2Face-3D provides audio-driven facial-animation tooling. GPU throughput depends on hardware; you also supply conversation logic, rendering or streaming infrastructure, and appropriate model licenses.