Real-Time Avatar Providers Compared (Late 2026): HeyGen LiveAvatar, Tavus, Anam, Simli, Beyond Presence and Self-Hosting
October 8, 2026 · 6 min read · Ortavox Team
A verified comparison of the real-time digital-human landscape in late 2026: what each provider actually sells (full stack or avatar-only), published per-minute pricing, concurrency, LiveKit plugin support and licensing, plus where a self-hosted renderer like OrtaHuman fits.
Two years ago a real-time avatar meant one vendor and a waiting list. In late 2026 the LiveKit Agents documentation alone lists sixteen avatar providers with a plugin, and prices range from under a cent a minute for a rendering layer to a third of a dollar a minute for a complete conversational stack. This guide sorts the landscape by what you are actually buying, with pricing taken from vendor pages at the time of writing.
Disclosure: written by the Ortavox team, which ships its own avatar layer (OrtaHuman). Every third-party figure below comes from the vendor's public pricing page or documentation as retrieved in October 2026. Where we could not retrieve a published number, the cell says so.
First decide what you are buying
Avatar products fall into two shapes, and comparing prices across the two is meaningless.
- Full-stack conversational video: the vendor runs STT, the LLM, TTS, the avatar and the WebRTC transport. You give it a persona and a knowledge base. Tavus CVI and HeyGen LiveAvatar's Full Mode are this shape. Simple to start, hard to customise, and your conversation logic lives inside the vendor.
- Avatar-only rendering: you run the voice agent and stream its audio to the vendor, which returns synchronised video. HeyGen's Avatar Only mode, Simli, Anam, Beyond Presence's speech-to-video product and every LiveKit avatar plugin are this shape. Cheaper per minute, and your agent stays yours.
Ortavox sits in the second camp by construction: the OrtaHuman renderer consumes the agent's TTS audio and publishes video, and the agent, tools and knowledge remain on the Ortavox engine.
Comparison matrix
Provider | Shape | Published entry plan | Per-minute (published) | Concurrency | LiveKit plugin |
|---|---|---|---|---|---|
HeyGen LiveAvatar | Both (Full Mode / Avatar Only) | Essential $99/mo, 1,100 credits | Full Mode 2 credits/min, Avatar Only 1 credit/min; enterprise "down to $0.01/min" | Unlimited on paid plans (per vendor) | Yes (Python) |
Tavus CVI | Full stack | Builder $59/mo, 175 min | $0.35 overage on Builder, $0.31 Growth, $0.26 Business | 3 / 10 / 15 by plan | Yes (Python, Node) |
Anam | Avatar-only (API/SDK) | Starter $12/mo, 50 min | $0.16 overage on Starter down to $0.11 on Professional; enterprise from $0.04 | 1 / 3 / 5 / 10 by plan | Yes (Python, Node) |
Simli | Avatar-only rendering | Hobby $10/mo, 1,000 min | About $0.01/min on paid plans; enterprise "as low as $0.005" | 2 / 10 / 50 by plan | Yes (Python) |
Beyond Presence | Both (speech-to-video / managed video agents) | Starter EUR 49/mo, 14K credits | Speech-to-video EUR 0.175 down to 0.0875; video agents EUR 0.35 down to 0.175; enterprise to EUR 0.03 | 10 / 25 / 50 by plan | Yes (Python, Node) |
D-ID Agents | Full stack | not publicly documented (pricing page not retrievable) | not publicly documented | not publicly documented | Yes (Python, Node) |
Ortavox OrtaHuman | Avatar-only, self-hostable | Beta; pricing on request | Not yet published | Planned per GPU pool | Native (built on the LiveKit avatar model) |
Notes on the matrix. HeyGen's LiveAvatar page lists credits per plan but not a dollar-per-minute rate; dividing the Business plan ($475 for 6,000 credits) gives roughly $0.08 per credit, so about $0.08 per minute in Avatar Only mode and $0.16 in Full Mode, before overage. Tavus rounds usage to six seconds with a 30-second minimum per conversation. Beyond Presence prices in euros and charges 50 credits per minute for speech-to-video and 100 for managed agents. D-ID's main pricing page returned only navigation text when we fetched it; its API pricing lives on a separate page we could not retrieve, so we have left those cells blank rather than copy third-party numbers.
Latency and quality: what is actually published
Most avatar vendors do not publish latency figures, and the ones that do publish a single number that is hard to compare. HeyGen's LiveAvatar page states a median time to first frame under 300 ms and 99.99% API uptime. Simli's pricing page makes no latency claim at all, only a 99.95% uptime SLA on enterprise. Tavus, Anam and Beyond Presence do not state a first-frame figure on their pricing pages. Treat vendor latency numbers as the renderer's contribution only; the end-to-end figure your user feels adds your STT, LLM and TTS on top.
Quality is likewise not something a spec sheet captures. The variables that matter in practice are lip-sync accuracy on fast consonants, behaviour during long silences (does the face idle naturally or freeze), how cleanly the mouth stops on an interruption, and whether a custom avatar made from your own footage looks like the person. The only honest evaluation is a side-by-side with your own script and your own TTS voice.
The LiveKit avatar model as the common interface
The reason the avatar-only shape has become the default for developers is that LiveKit Agents standardised it. An AvatarSession joins the room as a separate participant, the agent's audio output is redirected to the avatar worker over a data stream, and the worker publishes synchronised audio and video that front-ends recognise through a publish-on-behalf attribute. The documentation lists Anam, Avatario, AvatarTalk, Beyond Presence, bitHuman, D-ID, Keyframe, LemonSlice, LiveAvatar, Protoface, Runway, Simli, Spatius, Synthesia, Tavus and TruGen as plugins, all with Python support and seven with Node.js support.
For a buyer this means two things. Switching avatar vendors is a one-plugin change if your agent is on LiveKit. And a self-hosted renderer can implement the same interface, which is what Ortavox does: the OrtaHuman worker speaks the same AvatarSession and data-stream protocol, so the console, the React front-end and the agent do not know whether the face is rendered by a hosted API or by a GPU in your own rack.
Licensing and data
Question | Hosted avatar APIs | Self-hosted renderer (OrtaHuman) |
|---|---|---|
Where the video is rendered | Vendor's cloud | Your GPU pool or Ortavox-managed GPUs |
Where source footage lives | Uploaded to the vendor to create a custom avatar | Private object storage under your tenant, or on-premise |
Custom avatar rights | Vendor terms; usually per-avatar slots or fees (HeyGen lists add-on custom avatar slots; Simli lists custom avatars per month by plan) | Your approved source material; no per-avatar vendor fee |
Model licensing | Opaque; included in the per-minute price | Open lip-sync pipeline; model licences reviewed per component |
Session caps | Common: Anam 3-minute conversations on Free up to 2 hours on Growth; Tavus 5 to 60 minutes by plan; HeyGen 2 to 60 minutes by plan | Set by you |
Session-length caps are easy to miss and bite in kiosk and reception deployments where a conversation can run long. Check the cap on the plan you intend to buy, not the headline.
How to choose
- Prototype in a week with no agent of your own: Tavus or HeyGen Full Mode. Accept that the conversation logic is theirs.
- You already have a voice agent and want the cheapest face: Simli, then Anam or HeyGen Avatar Only, compared on your own footage.
- You bill in euros and want a managed agent option on the same account: Beyond Presence.
- Video or footage cannot leave your network, or you want avatar cost to be a GPU line item rather than a per-minute fee at volume: a self-hosted renderer. That is the case OrtaHuman is built for, and the beta is open to teams that want to measure it.
Sources
- LiveKit Agents: Virtual avatar models overview — https://docs.livekit.io/agents/models/avatar/
- HeyGen LiveAvatar — https://www.liveavatar.com/
- Tavus plans and pricing — https://www.tavus.io/pricing
- Anam pricing — https://anam.ai/pricing
- Simli pricing — https://www.simli.com/pricing
- Beyond Presence pricing — https://www.beyondpresence.ai/pricing
- D-ID pricing — https://www.d-id.com/pricing/
- Ortavox OrtaHuman — https://ortavox.ai/ortahuman