Skip to content

Images, video, voice and avatars from your code

One API and an MCP server over Seedance, Kling, Gemini Omni, FLUX, Nano Banana, ElevenLabs and the other models in the catalog. Prepaid in dollars, priced per call, every job itemised.

How one call runs

Six steps from your code to the file. Pick one to see the real request or response at that point.

  1. Step 1 of 6

    Your code or your agent

    Anything that can send an HTTPS request: a script, your backend, a workflow tool. An agent goes through the MCP server, which calls the same routes with the same key.

    • No SDK to install: JSON in, JSON out.
    • The example asks for a 4:5 product image.
    curl https://aitachyon.com/api/images/generate \
      -H "Authorization: Bearer ait_your_token" \
      -H "Content-Type: application/json" \
      -d '{
        "projectId": "<project id>",
        "prompt": "product on linen, soft daylight",
        "style": "Product Shot",
        "width": 1024,
        "height": 1280
      }'

First call in a minute

Four steps, each one a request you can paste.

  1. 1

    Create a key

    In Settings, under API tokens. The key starts with ait_, is shown once and works for the API and the MCP server alike.

  2. 2

    Check it

    GET /api/me is free. It says how you pay, your balance and your projects; most calls need a projectId from that list.

  3. 3

    Make the first call

    POST a prompt to /api/images/generate. The image comes back in the same response, with a signed link.

  4. 4

    Fetch it again

    GET /api/generations/{ref} returns any item in any status, with a freshly signed link.

Calls draw on a prepaid balance: top up from $10 in Settings, under Pay as you go.

# HTTP
Authorization: Bearer ait_your_token

# curl
-H "Authorization: Bearer ait_your_token"

# Node (fetch)
headers: { Authorization: "Bearer ait_your_token" }

# Python (requests)
headers={"Authorization": "Bearer ait_your_token"}

Video takes a minute: poll for it

An image comes back in the response. A video answers at once with a task to poll, and is charged when it starts.

  1. Your code

    Submit

    POST /api/scenes/generate with the prompt, the engine and the length.

    curl https://aitachyon.com/api/scenes/generate \
      -H "Authorization: Bearer ait_your_token" \
      -H "Content-Type: application/json" \
      -d '{
        "projectId": "<project id>",
        "prompt": "the product turning slowly on a kitchen counter",
        "engine": "seedance-2.5",
        "duration": 5,
        "ratio": "16:9"
      }'

    Aitachyon

    Accepted, charged

    The answer comes at once: the scene id, the task and the pollUrl to call.

    {
      "scene": {
        "id": "0c7d3b9e-2f41-4a8e-9b6d-71e5a2c48f03",
        "taskId": "<task id>",
        "status": "processing",
        "mode": "hd",
        "engine": "seedance-2.5",
        "resolution": "720p",
        "audio": false,
        "creditsCharged": 65,
        "pollUrl": "/api/scenes/status/<task id>"
      }
    }
  2. Your code

    Poll

    GET the pollUrl every few seconds.

    curl https://aitachyon.com/api/scenes/status/<task id> \
      -H "Authorization: Bearer ait_your_token"

    Aitachyon

    Processing, then completed

    While it renders: processing. The poll that sees it finished stores the clip and returns a signed link.

    {
      "status": "processing"
    }
  3. Your code

    Any time later

    The scene's ref is scn_ and its id: GET /api/generations/{ref} returns it with a fresh link.

    curl https://aitachyon.com/api/generations/scn_0c7d3b9e-2f41-4a8e-9b6d-71e5a2c48f03 \
      -H "Authorization: Bearer ait_your_token"

    Aitachyon

    {
      "ref": "scn_0c7d3b9e-2f41-4a8e-9b6d-71e5a2c48f03",
      "type": "scene",
      "id": "0c7d3b9e-2f41-4a8e-9b6d-71e5a2c48f03",
      "project_id": "8b1f2c4e-6d3a-4f5b-9c7e-2a1d0e9f8b7c",
      "status": "completed",
      "model": "seedance-2.5",
      "mode": "hd",
      "prompt": "<the directed prompt the model received>",
      "ratio": "16:9",
      "duration_seconds": 5,
      "credits_charged": 65,
      "url": "https://<storage>/scenes/<you>/<project>/0c7d3b9e-2f41-4a8e-9b6d-71e5a2c48f03.mp4?X-Amz-Expires=3600&X-Amz-Signature=<signature>",
      "error": null,
      "created_at": "2026-11-01T09:30:12.000Z"
    }

If it fails

The poll answers failed, and the charge comes back on its own ledger line, linked to the scene. There is nothing to claim.

GET https://aitachyon.com/api/wallet

{
  "ledger": [
    {
      "at": "2026-11-01T09:31:40.000Z",
      "kind": "refund",
      "amount_usd": 2.312,
      "balance_after_usd": 10,
      "action": "scene_generation",
      "model": "seedance-2.5",
      "credits": null,
      "provider_cost_usd": null,
      "markup_pct": null,
      "estimated": false,
      "note": "scene_failed:0c7d3b9e-2f41-4a8e-9b6d-71e5a2c48f03"
    },
    {
      "at": "2026-11-01T09:30:12.000Z",
      "kind": "charge",
      "amount_usd": -2.312,
      "balance_after_usd": 7.69,
      "action": "scene_generation",
      "model": "seedance-2.5",
      "credits": 65,
      "provider_cost_usd": 1.156,
      "markup_pct": 100,
      "estimated": false,
      "note": null
    }
  ]
}

Example values. A real ledger line carries the measured cost of your own call.

When a call is refused

Three answers to handle. None of them charges anything.

  • 401Nothing charged

    Not authenticated

    The key is missing, mistyped or revoked. GET /api/me says which: no_credentials, malformed_token or invalid_token.

    {
      "error": "Not authenticated."
    }
  • 402Nothing charged

    Payment required

    The balance does not cover the call, or the key reached its spend cap. The body says which and how much is left: top up or raise the cap, then run it again.

    {
      "error": "Insufficient balance: $0.04 left, not enough for this job. Top up at https://aitachyon.com/settings or POST /api/wallet/topup.",
      "code": "INSUFFICIENT_BALANCE",
      "balance_usd": 0.04
    }
  • 503Nothing charged

    Provider unavailable

    The service behind the model refused on our side. Pick another provider, or try again in a little while.

    {
      "error": "OpenAI images (gpt-image-2) are temporarily unavailable on our side. Use provider \"nanobanana\" or \"recraft\" in the meantime; nothing was charged.",
      "code": "PROVIDER_UNAVAILABLE"
    }

What a call costs

Prepaid prices. A range means the price depends on something only the run decides, like output length. You are always charged what the job actually cost.

Every model and unit, machine-readable: GET /api/pricing · the full reference: /docs/api

  • ImageSeedream 4.0$0.090
  • ImageNano Banana$0.10
  • ImageFLUX.2 [pro]$0.090 – $0.14
  • Video, 5 secondsSeedance 2.5, 480p$1.03
  • Video, 5 secondsSeedance 2.5, 720p$2.31
  • Video, 5 secondsKling v3$0.84
  • Voiceover, ~450 charactersElevenLabs$0.11
  • Music trackElevenLabs Music$0.030 – $0.22

The labs behind the models

Seedance, Kling, Nano Banana, FLUX, gpt-image-2, ElevenLabs and the rest: models from 15 labs, on one key.

Every model
  • ByteDanceSeedance 2.5, Seedream
  • KuaishouKling v3
  • GoogleGemini Omni 1.2 Flash, Nano Banana
  • AlibabaWan 3
  • MiniMaxHailuo 02
  • xAIGrok Imagine
  • Black Forest LabsFLUX.2 [pro]
  • OpenAIgpt-image-2
  • RecraftRecraft V4
  • ElevenLabsEleven v4 voices, Eleven Music
  • Fish AudioVoice cloning
  • AnthropicClaude Opus, Claude Sonnet
  • TripoTripo 3D
  • MicrosoftTRELLIS
  • TencentHunyuan3D

Or let your agent drive

The hosted MCP server gives Claude Code, Cursor and other agents the same calls as tools, on the same key and the same ledger. One command adds it.

Set up MCP

Works with

  • Claude Code
  • Cursor
  • Claude Desktop
  • claude.ai
  • ChatGPT
  • Any MCP client

claude mcp add --transport http aitachyon https://aitachyon.com/api/mcp

https://aitachyon.com/api/mcp

Building for your own customers?

Give each customer their own key with a spend cap, read what each one spent, and charge your own price. Apply for a wholesale rate.

See the white-label API