Media tools for your agent, with a price on every call.
One API key gives an agent image, video, voice, music and lip-sync. Every call can be priced before it runs, and caps on the key and the project stop an agent that loops. Base model cost plus a flat 20% platform fee, shown before you spend.
- 144 models
- Quote before every call
- Caps per key and per project
- Hosted MCP server
- Balance never expires
An agent with a credit card is a budget problem.
Agents retry, branch and loop. If the tool behind them bills from a monthly allowance or shows the cost after the fact, you find out what a run cost when the allowance is gone.
- A separate vendor and key for images, video, voice and music
- Cost visible after the call, in a plan's credits
- One account-wide limit, so one runaway job drains everything
- A plan sized for your busiest month
- One key for image, video, voice, music and lip-sync
- POST /v1/estimate returns the dollar price before the agent commits
- A per-request cap and a total cap on each key, plus a monthly cap per project
- A prepaid balance with no plan. Quiet months cost nothing
What the agent can call
Text to speech with a 168-voice library, speech to text, dubbing and voice changing.
Voice models →Songs with lyrics and instrumentals from the Suno and Mureka families. Suno Generate Music is $0.072/track.
Music API →Turn a face and a line of text, or a face and an audio file, into a talking video in one call.
Lip-sync API →Brief to storyboard to rendered video, with the render quoted before it starts.
AI Director →Quote, then run
The agent asks what a call costs, compares it with its budget, and only then generates. The same key works over REST and over the hosted MCP server.
# 1. Price the call. Nothing is charged.
curl -X POST https://api.fairstack.ai/v1/estimate \
-H "Authorization: Bearer $FAIRSTACK_API_KEY" \
-H "Content-Type: application/json" \
-d '{"modality":"video","model":"seedance-2-fast","duration_sec":10}'
# 2. Run it. A call over the key's per-request cap is refused, not billed.
curl -X POST https://api.fairstack.ai/v1/video/generate \
-H "Authorization: Bearer $FAIRSTACK_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"seedance-2-fast","prompt":"A paper boat in the rain","duration_sec":10}'What a call costs
A few of the models agents call most, at their default settings.
Live catalog prices: base model cost plus a flat 20% platform fee. Every call is priced before it runs. All models
Set it up once
- 01Make a project and a key
Give each agent its own key. Set a per-request cap and a total cap on the key, and a monthly cap on the project.
- 02Connect it
Call the REST API, use the TypeScript or Python SDK, or add the hosted MCP server to Claude, Cursor or Codex.
- 03Let it work
The agent quotes, runs and polls. Spend per key is in the usage API, so you can see which agent spent what.
What an agent needs from a media tool
Billing terms, not unit prices. Unit prices depend on model, mode and resolution.
| FairStack | Typical creator plan | Typical API gateway | |
|---|---|---|---|
| Pricing | Prepaid balance, no plan | Monthly plan with a credit allowance | Pay as you go |
| Price before the call | Yes, in dollars, on every call | Credits per action | List price per model |
| Image, video and voice | Yes | Yes | Yes |
| Web studio on the same balance | Yes | Yes | Rarely |
| Caps | Per request, per key, per project | Plan ceiling | Account limit |
| Unused balance | Never expires | Monthly allowance resets | Often expires after a year |
Frequently asked questions
How does an agent know what a call will cost? +
It calls POST /v1/estimate with the model and settings, or sends the generation request with confirm set to false. Both return the dollar price without charging anything. The price is base model cost plus a flat 20% platform fee.
What stops an agent from spending my whole balance? +
Each API key can carry a per-request cap and a total cap, and each project can carry a monthly cap. A call that would go over is refused before it runs. Storyboard renders can also carry their own budget, and a render quoted above it is refused.
Does FairStack have an MCP server? +
Yes. A hosted MCP server exposes generation, the model catalog, cost estimates and your asset library as tools. It uses your normal API key, and generations through it cost the same as over REST.
Do I need a plan or a subscription? +
No. FairStack has no plans. You add funds to one balance, $1 minimum, and each generation is billed at base model cost plus a flat 20% platform fee. Your balance never expires.
Which modalities can one key call? +
Image, video, voice (text to speech, speech to text, dubbing), music and lip-sync, across 144 models. 3D generation is available in the web app.
Still have questions? We're here to help.
Keep reading
- OpenRouter FAQ (expiry) (read 2026-10-31)
- Replicate prepaid credit (read 2026-10-31)
- fal terms, section 9(a) (read 2026-10-31)
- Higgsfield: what is the Higgsfield API (read 2026-10-31)
- WaveSpeedAI: how pricing works (read 2026-10-31)