Pricing & credits
How bitHuman bills — credits per active minute by model and surface, one-time creation costs, plans and concurrency, offline licensing, and how to check your balance.
How billing works
bitHuman bills in credits consumed per active minute of avatar runtime. Audio-only mode (the Swift SDK without an attached avatar) is unmetered. Plans top up credits monthly; overage is pay-as-you-go.
This page is the single source for every billing number on the platform — the model guides and API pages link back here.
Grab a free dev key at bithuman.ai → Developer — it lands in your inbox in seconds with the free tier attached.
Serving — credits per active minute
| Model | Cloud | Self-hosted |
|---|---|---|
Essence 2 (essence-2) | 4 credits/min | 2 credits/min |
Essence 2 Max (essence-2-max) | 8 credits/min | 4 credits/min |
Expression 2 (expression-2) | 4 credits/min | 2 credits/min |
Essence 1 (essence-1) | 2 credits/min | 1 credit/min |
Expression 1 (expression-1) | 4 credits/min | 2 credits/min |
Self-hosted serving is half the cloud rate across the board, and on-device serving (the Swift SDK) bills at the self-hosted rate. A “credit minute” is wall-clock time the engine is actively producing frames (on-device, the wall-clock between chat.start() and chat.stop() with an avatar attached) — idle, paused, or disconnected time isn’t billed. The second-generation models are launching July 10, 2026 — rollout in progress.
Managed conversational agents bill on top of avatar serving:
| Surface | Rate |
|---|---|
| Managed agent — voice chat | 10 credits/min |
| Managed agent — camera chat (vision on) | 30 credits/min |
Two modes are always free: audio-only Swift SDK use (no avatar attached — fully offline, no metering) and BITHUMAN_UNMETERED=1 dev mode (skips auth and heartbeat — development and parity testing only, not licensed for production).
Creation & generation — one-time credits
| Action | Credits (one-time) | Notes |
|---|---|---|
Agent creation — first generation (essence-1, expression-1) | 250 | Per .imx built from your portrait image |
Agent creation — essence-2 (combined) | 500 | One charge covers both Essence 2 models — Essence 2 and Essence 2 Max, trained on-create from your portrait image; pick the model at launch |
Agent creation — expression-2 | 2000 | Fully generative engine trained on-create from your portrait image — its per-identity training costs more to run, priced accordingly |
Agent creation — auto | 500 or 2000 | The platform classifies your input — a photorealistic person routes to essence-2 (500), a cartoon / animal / creature to expression-2 (2000); you’re billed the routed model’s rate |
Model add (POST /v1/agent/{code}/models) | same per-model rates | Add a model to an existing agent; adding Expression 1 is free (instant, nothing trained) |
| Dynamics generation | 250 | Per gesture / movement set generated for an agent |
| Book creation | 250 | Per illustrated book generated from a prompt |
Talking video — per minute of output
Talking-video renders bill per minute of finished output, rounded up (minimum one minute). A failed render is automatically refunded.
| Model | Per minute of output |
|---|---|
essence-2 | 4 credits |
essence-2-max | 8 credits |
expression-2 | 4 credits |
essence-1 | 2 credits |
expression-1 | 4 credits |
Free tier
- 99 credits / month at signup, no credit card required.
- Good for roughly 50 minutes of cloud Essence 1 serving, or about 25 minutes of the second-generation models.
- Resets monthly. Unused credits don’t roll over.
Plans
| Plan | Monthly | Yearly | Credits / month | Concurrent sessions |
|---|---|---|---|---|
| Free | $0 | — | 99 | 1 |
| Creator | $20 | $204 | 1,800 | 3 |
| Pro | $99 | $1,010 | 10,000 | 10 |
| Business | $299 | $2,990 | 50,000 | 50 |
| Enterprise | $999 | $9,990 | 250,000 | 200 |
| Custom | Contact sales | — | Volume / on-prem | Unlimited |
Annual plans bill 12× the monthly credits up front and save up to ~17% (about two months free on Business and Enterprise; ~15% on Creator and Pro) — choose monthly or annual at checkout. Custom covers volume, on-prem / air-gapped deployment, and bespoke SLAs beyond Enterprise: talk to sales.
Concurrent sessions are a plan entitlement — the number of live avatar sessions your account can run at once. Enforcement is rolling out: when limits apply, a session past your plan’s cap is refused with 403 CONCURRENCY_LIMIT_REACHED rather than degrading running sessions. Details in Rate limits & concurrency.
Current pricing and your live balance are in the bitHuman dashboard — the credit balance is on the top navigation bar.
Offline licensing — coming soon
Self-hosted serving today authenticates online (a once-per-minute billing heartbeat). Offline licensing — running models fully disconnected, with no heartbeat — is coming soon, unlocked by tier starting at Pro:
| Offline package | Models | Annual commitment | Offline credits included |
|---|---|---|---|
| Pro | Essence 1 + Expression 1 | from 60,000 credits/yr | 60,000 |
| Business | Essence 2 + Expression 2 | $999/yr | 120,000 |
| Enterprise | adds Essence 2 Max | $1,999/yr | 240,000 |
Offline serving consumes the included credits at half the equivalent cloud rate — Essence 1 at 1 credit/min and Expression 1 at 2; Essence 2 and Expression 2 at 2; Essence 2 Max at 4. Entitlements are delivered as per-device, per-model signed credit bundles: minted once while the device is online, then valid with no further connectivity until the credits are consumed. These packages are separate from the monthly plans above. Talk to sales to get on the early-access list.
Top-up credits
Need more before your next reset? Top up any time at $1 = 100 credits. Top-up credits never expire — only plan credits reset at the end of each billing period (unused plan credits don’t roll over). Top-ups stack on your plan balance and are spent after plan credits.
Metered vs unmetered
| Mode | What it means | Auth |
|---|---|---|
| Metered (default) | Your BITHUMAN_API_SECRET exchanges for a runtime token; a heartbeat fires once per minute while frames are flowing. Both cloud and self-hosted run this way. | BITHUMAN_API_SECRET (server) / BITHUMAN_API_KEY (Swift) |
| Unmetered dev mode | BITHUMAN_UNMETERED=1 skips auth + heartbeat entirely. For local dev, CI, and parity work — not licensed for production. | none |
| Audio-only | Swift SDK with no avatar config attached. Fully offline, never reaches the auth endpoint. | none |
The CLI, Python SDK, and Docker container all honour BITHUMAN_UNMETERED=1. The Swift SDK has the same escape hatch via its unmetered initializer.
How metering works
Server-side surfaces (cloud, self-hosted Python, self-hosted GPU)
The Python SDK and Docker container exchange a BITHUMAN_API_SECRET for a short-lived runtime token, then heartbeat back to api.bithuman.ai once per minute while the engine is generating frames. Each heartbeat increments your usage counter.
On-device surface (Swift SDK)
The Swift SDK requests a runtime token once on chat.start() (sync — bad keys fail fast with VoiceChatError.authenticationFailed), then heartbeats once per minute while the avatar is attached. Audio-only mode doesn’t authenticate or heartbeat at all. If the device loses connectivity mid-session, the SDK has a 5-minute offline grace period before it surfaces a billing error and pauses the avatar.
Check your balance
curl https://api.bithuman.ai/v2/credit-summaries \
-H "api-secret: $BITHUMAN_API_SECRET"Response:
{
"success": true,
"data": {
"user_id": "1f09e2f0-bffd-4201-a3e7-e98a9e432e67",
"balance": 5240,
"plan_credits": 240,
"topup_credits": 5000,
"is_enterprise": false,
"minutes_estimate": {
"voice_chat": 524,
"camera_chat": 174,
"essence_cloud": 2620,
"essence_self_hosted": 5240,
"expression_cloud": 1310,
"expression_self_hosted": 2620
}
}
}The minutes_estimate keys map directly to the two avatar model families: essence_* = Essence, expression_* = Expression (cloud = bitHuman-hosted, self_hosted = your hardware). voice_chat / camera_chat are managed cloud conversational-agent estimates, not avatar models.
What’s NOT billed
- Source code, SDK installs, documentation — free.
- Audio-only Swift SDK use — voice chat with no avatar attached is unmetered and fully offline.
- Idle time — the engine has to be actively producing frames. Stopped or paused sessions don’t accrue.
- Failed auth — bad keys fail fast and don’t burn credits.
- Failed creations and renders — automatically refunded.
- Model weights —
.imxand Expression weight downloads are free; only active runtime minutes count.
FAQ
Does the on-device Swift SDK work without an internet connection?
Audio-only mode is fully offline. Avatar mode authenticates once on chat.start() and heartbeats once per minute — with a 5-minute offline grace window after the last successful heartbeat. After that, the avatar pauses until connectivity returns. For fully disconnected deployments, see Offline licensing.
What if I run out of credits mid-session?
The current heartbeat finishes, then subsequent heartbeats fail. The Python / Docker engines stop emitting frames; the Swift SDK surfaces a VoiceChatError.authenticationFailed and the avatar pauses. Top up credits to resume.
Can I have multiple concurrent sessions?
Yes. Each session bills independently while it’s actively generating frames. Your plan sets the concurrent-session entitlement — see the plans table (enforcement is rolling out). Self-hosted deployments are additionally bounded by your own hardware.
Next steps
- Billing API — credit summaries over REST.
- Building avatars — create an agent to run.
- Essence 2 & Expression 2 — the second-generation models and how to choose.
- Rate limits & concurrency — request limits and the concurrency contract.
- Deploy via LiveKit — the cloud-metered path.