Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.bithuman.ai/llms.txt

Use this file to discover all available pages before exploring further.

bitHuman bills in credits consumed per active minute of avatar runtime. Audio-only mode (Swift SDK without an attached avatar) is unmetered. Plans top up credits monthly; overage is pay-as-you-go.

​
At a glance

SurfaceModelCostNotes
bitHuman CloudEssence2 cr/minCPU rendering on bitHuman’s servers
bitHuman CloudExpression4 cr/minGPU rendering on bitHuman’s servers
Self-hosted (Python SDK)Essence1 cr/minCPU on your hardware
Self-hosted (GPU container)Expression2 cr/minNVIDIA GPU on your hardware
On-device (Swift SDK)Expression2 cr/minActive avatar minutes only
On-device (Swift SDK), audio-only—FreeNo avatar, no metering
Agent generation—250 cr (one-time)Per .imx model built from your photo / video
A “credit minute” is wall-clock time the engine is actively producing frames (or, on-device, the wall-clock time between chat.start() and chat.stop() with avatar attached). Idle / paused / disconnected time isn’t billed.

​
Free tier

  • 99 credits / month at signup, no credit card required.
  • Generates ~50 minutes of cloud Essence or ~25 minutes of cloud Expression.
  • Resets monthly. Unused credits don’t roll over.

​
Plans

PlanPriceCredits / monthConcurrent sessions
Free$0992
Creator$201,2005
Pro$997,50010
EnterpriseContactCustomCustom
Current pricing is shown in your bitHuman dashboard — the credit balance is on the top navigation bar.

​
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 calls https://api.bithuman.ai/v1/runtime-tokens/request 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 (no AvatarConfig) doesn’t authenticate or heartbeat at all — it’s fully offline and free. 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 -X GET https://api.bithuman.ai/v2/credit-summaries \
  -H "api-secret: $BITHUMAN_API_SECRET"
Response: 
{
  "balance": 1842,
  "plan": "creator",
  "renews_at": "2026-05-15T00:00:00Z",
  "estimated_minutes": {
    "essence_cloud": 921,
    "expression_cloud": 460,
    "expression_on_device": 921
  }
}
See /api-reference/credit-summaries for the full schema.

​
What’s NOT billed

  • Source code, SDK installs, documentation — free.
  • Audio-only Swift SDK use — voice chat with no avatar attached is unmetered and runs entirely 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.

​
FAQ

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.
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.
No. Weight downloads (the universal Expression bundle for the Swift SDK; per-agent .imx files for the Python SDK) are free. Only active runtime minutes count.
Yes, up to your plan’s concurrent-session limit. Each session bills independently while it’s actively generating frames.