Credits & Pricing
What operations cost, how to top up, and how spending is tracked
Overview
Credits are the organization's shared balance for generation work. Every member and every API key draws from the same pool. New organizations start with 1,000 credits; a low-balance warning appears under 50. Browsing, saving, playing existing audio, and managing content never cost credits — only generation does.
Cost Table
Speech generation
Speech generation is cost-based: Tonetta charges what the generation actually cost at the AI provider, times a surcharge, rounded up to the next credit (minimum 1 credit). One credit corresponds to one cent of pack price, so a generation that cost $0.03 at the provider charges 6 credits at the standard 2× surcharge. Short lines cost less than long ones — you pay for what you use. When a provider reports no cost for a generation, the flat rates below apply instead.
| Operation | Cost (credits) |
|---|---|
| TTS — external voice (dashboard, API, or live endpoint) | actual cost × surcharge (flat fallback: 5) |
| TTS — internal Voice Studio voice | actual cost × surcharge (flat fallback: 10) |
| Voice preview — external | actual cost × surcharge (flat fallback: 5) |
| Voice preview — internal | actual cost × surcharge (flat fallback: 10) |
Voice Studio
| Operation | Cost (credits) |
|---|---|
| Design voice previews (per run) | 25 |
| Save a designed voice | 25 |
| Clone a voice | 100 |
Sound effects
| Operation | Cost (credits) |
|---|---|
| SFX generation | Per second of requested audio (provider rate, with optional free seconds and a minimum) |
3D models
Model generation runs as a background job, so it is charged in two steps: a flat reservation when the job is submitted, then a settlement to the actual cost — the LLM tokens the agent used plus the compute minutes, times the surcharge — when it finishes. The difference comes back; the reservation is the most a job can cost. The amounts below are the base reservation per tier; each model in the catalog shows its own ≈ credits estimate (what a job on it typically settles to) and, for premium models, a larger reservation sized from that estimate — always in Tonetta credits, never the provider's prices.
| Operation | Base reservation (credits) |
|---|---|
| Generate — quick tier | 25 |
| Generate — detailed tier | 90 |
| Refine — quick tier | 20 |
| Refine — detailed tier | 70 |
Cache hits are free
Live endpoints only charge when audio is actually generated. Serving cached audio — including stale audio while a refresh runs — costs nothing.
Charging and Refunds
- •Charged up front, refunded on failure. Generation reserves its credits before work starts; if the provider errors or times out, the reservation is refunded automatically. A failed generation never costs anything.
- •Cost-based settlement. After a successful generation the reservation is reconciled against the provider's actual reported cost times the surcharge — you get the difference back when the generation was cheaper than the flat reservation.
- •3D model jobs. A failed job refunds the whole reservation (
…_refundin the audit log); a job you cancel is charged only for the cost incurred before it stopped; a job that stops reporting is failed and refunded automatically. The settlement row (…_adjustment) records the LLM and compute cost separately. - •Insufficient balance →
402. Dashboard flows show the shortfall; API responses use theinsufficient_tokenserror code. Live endpoints keep serving cached audio with atokenErrornote instead of failing. - •Provider fees are separate. External voice usage (e.g. ElevenLabs character quota) is billed by the provider to your own account; Tonetta credits cover Tonetta's processing.
Buying Credits
- 1Go to Purchase Credits (also linked from the Credits page).
- 2Pick a pack — Starter (500 credits) or Standard (1,000 credits) — and check out with Stripe.
- 3Credits are added the moment payment completes; the purchase appears in the audit log.
Safe checkout
Fulfilment is idempotent — a retried webhook or a refreshed success page can never double-credit a purchase.
Tracking Spend
The Credits page shows the current balance and the audit log: every add and spend with a timestamp, message, and balance after. Entries carry metadata — which operation, which live endpoint or generation, which project — so API-driven spending is attributable. Spends from API keys are recorded like any other, making it easy to see what an integration costs.
- •Watch for repeated
MISSgenerations from a live endpoint with a too-short TTL — lengthening the TTL is usually free money. - •Voice cloning (100) and design runs (25) are the big one-off costs; per-line TTS is cheap.
- •Scope API keys tightly (see API Keys & Scopes) so only intended integrations can spend.