Waterr AI Logo
ComparisonOctober 27, 2026

HeyGen alternatives for developers: 9 real-time avatar APIs compared (LiveAvatar, not video generation)

HeyGen's real-time product is now LiveAvatar, with its own billing, session caps and credit maths. Here are the alternatives that stream, priced per minute and dated 2026-10-27, plus what to check before you migrate.

Harshit SharmaFounder & CEO, Waterr AI

If you are leaving HeyGen's real-time product, you are leaving LiveAvatar, not the HeyGen video studio. Alternatives that stream a talking face to a browser include Anam, Simli, LemonSlice, Beyond Presence, Tavus, D-ID and Runway Characters. If you need the finished call (a transcript, a score against goals, a booking page) and not only a face, a meeting-lifecycle API such as Waterr is a different category. Published per-minute prices run from about $0.01 to $0.35. Everything below is dated 2026-10-27.

HeyGen's product is good. This is a map of what sits next to it, not a takedown.

The gap

You built a demo on HeyGen's streaming avatar. It worked. Then someone scheduled a 30-minute mock interview and the session died at minute 20, because that is what the Essential plan allows. Finance asked what a minute costs and the answer was "credits". Product asked for a score at the end, and you realised you were about to write that layer yourself.

That is a product boundary, not a defect. The question is which side of it your problem sits on.

Which "HeyGen" are you leaving?

Three things share the name.

ProductWhat it doesWhere it lives
HeyGen video generationRenders one-way avatar video from a script or prompt. Not real-time.heygen.com, developers.heygen.com
Streaming Avatar / Interactive Avatar API (legacy)The original real-time API. Its SDK (@heygen/streaming-avatar) still pulled 81,558 npm downloads between 2026-09-26 and 2026-10-25 per the npm API.docs.heygen.com
LiveAvatarThe renamed, standalone real-time product. HeyGen told its community about the rename on 6 Nov 2025 and says migration of every asset "is not guaranteed" (HeyGen blog). Own domain, docs, dashboard and credits that do not transfer from the studio.liveavatar.com, docs.liveavatar.com

LiveAvatar runs over WebRTC through LiveKit (liveavatar.com/llms.txt). It has two modes:

  • FULL burns 2 credits a minute. HeyGen runs the conversation stack and renders the face.
  • LITE burns 1 credit a minute. You bring the brain and push it text or audio, and HeyGen renders the face.

What LiveAvatar does ship: session transcripts (docs), memory for continuity across sessions, custom LLM, and interruption events. What I found no documentation for: scored analysis, goals, scheduling or booking, and stored recordings. HeyGen's own blog says LiveAvatar "streams over WebRTC without saving video."

Why developers look for alternatives

Four reasons keep coming up. Each is sourced to HeyGen's own pages or to a clearly labelled third party.

1. Session caps. From liveavatar.com: Free is 2 minutes per session, Essential ($99) is 20, Business ($475) is 60. A 45-minute sales role-play or interview is a Business-plan feature.

2. Credit maths. One credit pool, two burn rates, and overage priced at $0.095 per credit on Essential and $0.09 on Business in the credits FAQ. Credits are separate from your HeyGen studio plan, so you may pay two bills.

3. API pace. The JS SDK was at v0.0.19 on 16 September, image avatars are alpha, and per-key rate limits of 10 requests a second arrived on 2026-09-01 (changelog). A product moving fast is good, and your integration moves with it.

4. No scored output. You get the transcript. The rubric, the goals and the verdict are yours to build. HeyGen's own open-source sales agent sample does post-session summarising to a CRM, which shows where the demand is, but that is sample code and not an API feature.

On latency, be careful. HeyGen markets "median time to first frame under 300ms" on liveavatar.com. Anam published a vendor-run benchmark that reports a slower reply for HeyGen. Anam is a competitor grading its own test, and "first frame" and "reply" are different clocks. Test your own path before you believe either number.

The three kinds of alternative

The market splits into complete platforms and avatar layers. There is a third kind.

  1. Avatar layer. You run the STT, LLM and TTS. The vendor renders a face and streams it. Anam, Simli, LemonSlice, Beyond Presence, D-ID, Runway Characters. This is LiveAvatar's LITE mode, as a product.
  2. Complete real-time platform. The vendor runs the conversation and the face. Tavus is the best-known. This is LiveAvatar's FULL mode, as a product. Our Tavus alternatives post covers this tier at length, and the conversational video API explainer covers the definition.
  3. Meeting-lifecycle API. The unit is a completed meeting: scenario in, join link out, then recording, transcript and a scored evaluation back via API and webhook. The face is optional. Waterr sits here. See AI meeting agent API.

Pick by the thing your user receives. A face in a widget is category 1 or 2. A finished, scored conversation is category 3.

Comparison table

"Per minute" is what the vendor publishes, not a like-for-like cost: some minutes include STT, LLM and TTS and some render only. "Not documented" means I found no page that says so, not that the feature is absent. Sources are listed at the end.

VendorPublished $ per minuteFree tierMax sessionConcurrencyBYO LLMTranscriptScored analysisLiveKit / Pipecat
HeyGen LiveAvatarLITE $0.095 / $0.09, FULL $0.19 / $0.18 (Essential / Business overage, 1 and 2 credits a minute)10 credits/mo, 2-min cap2 / 20 / 60 min (Free / Essential / Business)1 free, unlimited paidYesYes (endpoint)Not documentedBoth
Anam$0.16 Starter, $0.14, $0.12, $0.11; Enterprise from $0.0430 min/mo, 3-min cap3 / 5 / 10 / 120 min by tier1 to 10, 100+ EnterpriseYesYes (API; none on zero-retention)Not documentedLiveKit yes; Pipecat listed, no guide
Simli$0.01 Hobby, $0.0095 Pro, $0.009 Scale50 min/mo, template avatarsNot documented1 / 2 / 10 / 50Yes (OpenAI-compatible endpoint)Not documentedNot documentedBoth
LemonSlice$0.22 extra min on StarterFeatured avatars only, no API30 min / 2 h / 24 h3 up to 100, 1,000+ EnterpriseYesNot documentedNot documentedBoth
Beyond PresenceEUR 0.35 / 0.20 / 0.175 (conversational agents); Enterprise from EUR 0.0320 min, 3-min cap3 min on Free, unlimited on paid1 / 10 / 25 / 50YesYes (API)Not documentedLiveKit yes; Pipecat not documented
TavusOverage $0.35 Builder, $0.31 Growth, $0.26 Business20 CVI min (page also shows 25), 1 concurrent5 / 10 / 15 / 60 min on one version of its pricing page1 / 1 / 3 / 10 / 15YesUtterance events and captionsBinary objectives, no rubric scoring (as of July 2026)Pipecat service
Runway Characters$0.20 plus $0.02 per sessionNot documentedNot documentedNot documentedYes (via LiveKit Agents)Not documentedNot documentedLiveKit Agents
D-IDNot publishedNot documentedNot documentedNot documentedYesYes (session data)Not documentedLiveKit yes; Pipecat not documented
Synthesia (interactive API)Not publishedNot documentedNot documentedNot documentedYesNot documentedNot documentedLiveKit Agents plugin
WaterrNot published; $1 = 1 credit, see pricing and the pricing calculator$5 one-time trial credit, 15-min calls, 30 days15 min / 1 h / 2 h / 4 h (Trial / Starter / Growth / Scale)1 / 3 / 10 / 30Platform-managed; your own keys on EnterpriseYesYes (goals and evaluations)None; hosts the meeting itself

Inside the plan, a credit costs less: Essential $99 for 1,100 credits is $0.09, Business $475 for 6,000 is $0.079. The per-minute figures in the table are overage rates, which is what you pay once the included credits run out.

Tavus shows two versions of its pricing page. One lists Free, Starter $22, Builder $59, Growth $397 and Business $975; the other lists Free, Starter $59, Growth $397 and Enterprise. Screenshot and date it before you quote it. Waterr's per-minute cost is a platform fee plus a voice stack plus add-ons such as the animated avatar, which is why no single figure is published.

The alternatives, one at a time

Anam

Quick summary. Avatar layer with the clearest published price ladder and a free API tier. Pros. $0.11 to $0.16 per minute on published plans, a free tier, custom LLM support, and 2-hour sessions on Growth and above. Cons. Free and Starter sessions are capped at 3 to 5 minutes. Its latency benchmark is self-run. Pick it if you want a LITE-mode replacement and plan to own the STT, LLM and scoring yourself. On heygen vs anam: Anam is cheaper per published minute and allows 2-hour sessions at its higher tiers, while HeyGen has no concurrency cap on paid plans.

Simli

Quick summary. The cheapest published face by a wide margin. Pros. $0.01 per minute on Hobby, 50 free minutes a month, LiveKit and Pipecat integrations. Cons. It renders the face, so the voice stack and scoring are yours. No session limit is documented. Pick it if you already run a voice agent and want to bolt a face on at the lowest cost.

LemonSlice

Quick summary. Long sessions and high concurrency ceilings. Pros. 30-minute calls on paid plans, up to 24 hours on Enterprise. Cons. The free tier has no API. Pick it if you need very long sessions. Test connect time first.

Beyond Presence

Quick summary. Pricing in euros, 10 concurrent sessions from the first paid tier. Pros. EUR 49 Starter, unlimited session length on paid plans, transcripts via API. Cons. The free tier is 3 minutes per session. Pick it if you want concurrency without enterprise sales.

Tavus

Quick summary. The full conversation platform, with the stack and the face from one vendor. Pros. Custom LLM, white label and compliance options at higher tiers, Pipecat service. Cons. Pricing varies across versions of its own page, so screenshot and date it. No rubric scoring documented as of July 2026 (comparison). Pick it if the face is the product. Our Tavus alternatives post goes deeper.

D-ID and Runway Characters

D-ID. Long-running avatar vendor with custom LLM support and session transcripts. It publishes no per-minute price. Pick it if you are already a customer.

Runway Characters. $0.20 a minute plus $0.02 per session, with a LiveKit Agents integration where you bring your own agent. No session limit is documented. Pick it if you already generate with Runway and run on LiveKit.

Waterr

Quick summary. A meeting API, not an avatar layer. You create a scenario and persona, send a join link, and read back the recording, transcript and a scored evaluation. The avatar is optional: a static face by default, animated in real time if you switch it on. Rendering is not our wedge. Pros. Goals and post-meeting evaluations, webhooks, scheduling and booking pages, embed tokens, MCP server, Claude Code and Codex skills. Cons. Platform-managed models on self-serve, with your own keys on Enterprise. No per-minute price is published. Pick it if the deliverable is a completed interview, role-play or feedback call. For the LiveKit comparison, see Waterr vs LiveKit.

Pricing maths for a real workload

Say 500 sessions a month, 20 minutes each, 10,000 minutes.

OptionPublished rateMonthly costNot included
HeyGen LITE, Business1 credit a minute, $0.09 overage$835 ($475 plan plus 4,000 overage credits)STT, LLM, TTS
Anam Professional$0.11 overage past 8,000 minutes$1,219 ($999 plus 2,000 minutes)STT, LLM, TTS
HeyGen FULL, Business2 credits a minute, $0.09 overage$1,735 ($475 plus 14,000 overage credits)Nothing, FULL runs the stack

The numbers are not comparable until you add what each minute leaves out. Write down what your minute must include, then price it. Waterr is not in this table because it publishes no per-minute figure.

Migration notes

What maps from LiveAvatar to each category.

LiveAvatar conceptAvatar layer (Anam, Simli, etc.)Complete platform (Tavus)Meeting-lifecycle API (Waterr)
FULL modeYou assemble STT, LLM, TTS yourselfClosest equivalentPersona plus scenario script
LITE modeDirect equivalent: push text or audio, get videon/an/a
Avatar IDsNot portable; re-pick from the new vendor's catalogRe-pickCatalog avatar assigned to a persona
Session token and keep-aliveNew session model per vendorPer vendorJoin link or embed token
Session transcript endpointAnam and Beyond Presence expose one; check the restUtterance events and captionsTranscripts endpoint
Memory summaryYours to buildCheck Tavus docsParticipant memory across sessions
Interruption eventsCheck each vendor's docsInterrupt and speaking-state eventsHandled in the meeting engine
CreditsPer-minute plansPlan minutesCredits, $1 = 1 credit

Do not port avatar IDs. And keep your LLM layer outside the avatar layer, so the next swap is a one-file change.

Pick HeyGen if

Be fair to the incumbent. Stay on LiveAvatar if:

  • You want 1080p. A custom avatar at 1080p is on the Business plan (liveavatar.com), and the preset library is 1080p.
  • You want no concurrency cap on paid plans, per liveavatar.com.
  • Your stack is LiveKit, Pipecat or Agora, all of which have documented HeyGen plugins.
  • Your sessions fit inside 20 or 60 minutes and you own the post-call layer already.

Frequently asked questions

How much does HeyGen cost per minute?

On LiveAvatar, LITE mode burns 1 credit a minute and FULL burns 2. Overage credits cost $0.095 on Essential and $0.09 on Business, so LITE is $0.095 or $0.09 a minute and FULL is $0.19 or $0.18. Video generation is priced separately.

Is there a free HeyGen alternative?

Several have free tiers. Simli gives 50 minutes a month, Anam 30 minutes, Tavus 20 to 25 CVI minutes, Beyond Presence 20 minutes, and Waterr a $5 one-time trial credit. HeyGen's own free plan is 10 credits a month with 2-minute sessions. Anam and Beyond Presence cap free sessions at 3 minutes.

Does HeyGen have an API?

Yes, more than one. The video generation API lives at developers.heygen.com. The real-time one is LiveAvatar (docs.liveavatar.com) with a JS SDK and LiveKit, Pipecat and Agora plugins. The older Streaming Avatar API still has an active SDK.

Is HeyGen's interactive avatar real-time?

Yes, through LiveAvatar over WebRTC. HeyGen claims a median time to first frame under 300ms. Anam's vendor-run benchmark reports slower replies. Neither is neutral. Measure on your own network and model.

What is the difference between LiveAvatar and HeyGen?

HeyGen is the video studio. LiveAvatar is the renamed Interactive Avatar real-time product, announced to the community on 6 Nov 2025, with its own site, docs, dashboard and credits that do not transfer between the two.

Which alternative gives a transcript and score?

HeyGen already gives a transcript through its API, and so do Anam, Beyond Presence and D-ID. A score is where they stop: I found no documented scored analysis at any of them, and Tavus documents binary objective completion. Waterr returns transcript, recording and a goal-scored evaluation through API and webhook.

Can I bring my own LLM?

On LiveAvatar, yes: custom LLM in FULL mode and full control in LITE. Anam, Simli, LemonSlice, Beyond Presence, Tavus, D-ID, Runway and Synthesia all document a bring-your-own-LLM path too. Waterr manages models on self-serve and allows your own keys on Enterprise.

What to do next

Decide what your user receives: a face in a widget, or a finished conversation. If it is a face, put Anam, Simli or Tavus through your own latency and connect-time tests this week, and price a real month of minutes with the other parts included. If it is a finished conversation, run the two-call quickstart and see whether the scored result saves you the layer you were about to write. Every figure here will drift, so recheck the vendor pages before you commit. For the full per-minute price table across vendors, see real-time avatar API pricing, which we update monthly.

If you remember one line: a face is a feature, and a finished conversation is a product.

Sources

All fetched 2026-10-27.

HeyGen alternativesLiveAvatar alternativeReal-time avatar APIInteractive avatarTavusAnamAI meeting API