Pricing explainer · 2026
The True Cost of AI Avatars, Decoded
For real-time, interactive AI avatars, the headline per-minute price is rarely what you pay — teaser rates, expiring credits, session caps and build costs all hide in the gap. Here's how to find the real number.
To compare real-time AI avatars honestly, ignore the headline rate and find the effective, all-inclusive cost — then check whether per-user memory is included. Once you do, the order is clear: for 50 hours of conversation a month you'd pay about $50 with Selvia AI, ~$480–600 with Anam/HeyGen/LemonSlice, ~$1,035 with Tavus, and ~$2,340 with D-ID.
A note on scope: this covers real-time, interactive avatars — the kind you pay for by the minute of live conversation. AI video generators (which render pre-recorded clips) price differently, by video, seat or output minute; see interactive avatar vs AI video generator for that distinction.
Five things that inflate the real cost
- 1.Teaser vs effective rates. Some advertised numbers are overage rates, not what a real plan costs. The effective rate is usually higher.
- 2.Expiring credits. Many platforms bill in monthly credits with no rollover, so minutes you don't use are simply forfeited.
- 3.Session caps & minimums. Limits on how long a single conversation can run, or monthly minimums, push up the real per-useful-minute cost.
- 4.What you have to build. If a rate covers only the avatar, you add the voice, language model and the engineering to wire it together. A bring-your-own-model option saves the platform fee but adds API cost and dev time; an API-only product (HeyGen's real-time avatar) means you build the whole surrounding app.
- 5.Per-user memory. The cost pricing pages never show: if memory isn't included, building an avatar that remembers returning customers is real, ongoing engineering you own.
How to read the price
| Platform | Advertised | Effective | Memory | The catch |
|---|---|---|---|---|
| Selvia AI | $0.03/min ($1/hr) | $0.03/min | Included | None — all-inclusive, pay for what you use |
| Anam | $0.11–0.16/min | ~$0.15–0.20/min | No | Advertised = overage rate; no rollover; session limits |
| LemonSlice | ~$0.14–0.19/min | ~$0.14–0.19/min | No | Sold as monthly credit buckets; no free tier |
| HeyGen | ~$0.20/min | ~$0.20/min + build | No | Real-time is API-only — you build the app; credits expire |
| Tavus | ~$0.32–0.37/min | ~$0.32–0.37/min | Marketed | Priciest all-inclusive — but it does include memory |
| D-ID | ~$0.78/min | ~$0.78/min | No | By far the most expensive real-time option |
Real-time / interactive rates as of June 2026; figures are approximate and change over time. “Effective” reflects typical paid-plan cost, not teaser or overage rates.
A worked decode: Anam
Anam is the clearest illustration of why headline rates mislead. Its advertised $0.11–0.16/min looks like one of the cheapest options — but that's the overage rate. On its Professional plan the effective cost works out closer to $0.20/min (~$12/hour), unused minutes don't roll over, and session limits cap how long a single conversation can run. It's a capable, very realistic avatar — but the real cost and usable scope are different from what the headline number suggests. (Anam is also a good avatar; this is about price transparency, not quality.)
Worked example: 50 hours of conversation a month
A single moderately busy assistant — roughly 100 minutes of live conversation a day. Here's the all-inclusive monthly cost at each platform's effective real-time rate:
| Platform | Per hour | 50 hrs / month | Notes |
|---|---|---|---|
| Selvia AI | $1 | $50 | All-inclusive; 60 min/mo free |
| LemonSlice | ~$9.60 | ~$480 | Credit buckets; no free tier |
| Anam | ~$12 | ~$600 | Effective rate; minutes don't roll over |
| HeyGen | ~$12 | ~$600 | Plus engineering to build the integration |
| Tavus | ~$21 | ~$1,035 | ~12× — but includes memory |
| D-ID | ~$47 | ~$2,340 | Most expensive by a wide margin |
Illustrative, based on effective real-time rates (June 2026) × 50 hours; actual costs depend on plan, usage pattern and add-ons. HeyGen's figure excludes the engineering to build the integration. Only Selvia AI and Tavus include per-user memory at these prices.
Our own numbers, in the open
In the spirit of the above, here is exactly what Selvia AI costs, with no asterisks: $1/hour ($0.03/minute), all-inclusive — the avatar, voice, language model and per-user memory are all in that rate. There's a render-only mode at $0.025/minute, a 60-minute monthly free tier, no expiring credits, and no session caps. The developer API is the same rate.
Why so low? It's structural, not a stripped-down product. Rendering the avatar is the expensive part of a real-time stack; the speech, language and memory layers add very little on top. Because we run lean, we can bundle all of it into one low rate instead of charging for each piece separately. Anam and Tavus render more photorealistic faces; our edge is price, automatic emotion and included memory.
See the price-ranked shortlist in the most affordable real-time AI avatars, or the full comparison in the 6 best real-time AI avatars of 2026.
Frequently asked questions
Why is real-time AI avatar pricing so hard to compare?
Because the headline per-minute rate rarely equals what you pay. Some advertised rates are overage rates, not the effective rate on a plan; many bill in monthly credits that expire; some cap session length; and some quote only the avatar, leaving you to add and pay for the voice, language model and memory yourself. To compare fairly, look at the effective rate for the same managed, all-inclusive setup — and check whether per-user memory is included.
What's the difference between an advertised and an effective rate?
An advertised rate is the number on the pricing page; an effective rate is what a real workload actually costs once you account for the plan you're on, expiring credits, and minimums. Anam is the clearest example: its advertised $0.11–0.16/min is the overage rate, while the effective rate on its Professional plan works out closer to $0.20/min — and unused minutes don't carry over.
What does a real-time AI avatar actually cost per month?
For 50 hours of conversation in a month, the all-inclusive cost is roughly: Selvia AI $50, LemonSlice ~$480, Anam ~$600, HeyGen ~$600 (plus engineering), Tavus ~$1,035, and D-ID ~$2,340. Selvia AI is the lowest because it's all-inclusive at $1/hour with nothing to add; the others are higher, and only Tavus includes per-user memory at its price.
Do cheaper-looking avatars include the voice and AI brain?
Sometimes, sometimes not. Several platforms offer a managed pipeline (voice + language model handled for you) or a bring-your-own-model option. If you choose bring-your-own, you add the language-model API cost plus the engineering to wire and maintain it. HeyGen's real-time avatar is API-only, so you always build the surrounding app yourself. Selvia AI's $0.03/min already includes the avatar, voice, brain and memory.
Why is per-user memory a hidden cost?
Because most real-time avatars don't include it, and pricing pages rarely mention it. If you want an avatar that recognizes returning customers and recalls past conversations, on most platforms that's a system you design, build and host yourself — real, ongoing engineering cost. Selvia AI includes per-user memory in the base price, and Tavus markets it; the others generally leave it to you.
Why is Selvia AI so much cheaper — is it lower quality?
No — the low price is structural, not a stripped-down product. Rendering the avatar is the expensive part of a real-time stack; the speech, language and memory layers add very little on top of that. Because Selvia AI is a lean, efficient operation, it can bundle all of those into one low, all-inclusive rate rather than charging for each separately. Anam and Tavus render more photorealistic faces; Selvia AI's edge is price, automatic emotion and included memory.
Pricing with no asterisks
All-inclusive real-time conversation with memory, $1/hour — start with 60 free minutes.
Book a Free DemoPricing based on publicly available information as of June 2026; figures are approximate and change over time. Check each vendor's site for current rates. All product names are trademarks of their respective owners; this is an independent analysis, not an endorsement.