Images

Sable exposes OpenAI-compatible image generation: point an OpenAI image client at the gateway and it works, metered per image, with a signed receipt on the response like every other billable call.

Availability

Image generation is off by default and config-gated. It is enabled only on a deployment where an image provider has been configured. Where it is not enabled, the model list comes back empty and a generation call returns a configuration error rather than a picture.

GET /v1/images/models is public and is the authoritative check for what, if anything, is enabled on this deployment. An empty data array means image generation is not available here.

# Public: what image models are enabled here (empty when off)
curl https://api.buildsable.com/v1/images/models

{
  "data": [
    { "id": "sable-image-…", "kind": "image", "price_usd_per_image": 0.04 }
  ]
}

Generating an image

Where a provider is enabled, POST /v1/images/generations takes the OpenAI shape and returns the OpenAI shape, plus a sable_receipt you can verify. It is key-authed: send a sk-sable_ API key as Authorization: Bearer.

curl https://api.buildsable.com/v1/images/generations \
-H "authorization: Bearer $SABLE_API_KEY" \
-H 'content-type: application/json' \
-d '{
  "model": "sable-image-…",
  "prompt": "a calm monochrome eclipse over an ink-black plane",
  "n": 1,
  "size": "1024x1024"
}'

{
  "created": 1756900000,
  "data": [{ "url": "https://…" }],
  "sable_receipt": {
    "receipt": "eyJ2Ijoz…",
    "signature": "0x4f8c…",
    "signer": "0xA1b2…9F"
  }
}

The receipt is metadata-only and verifies through the public POST /v1/receipts/verify, so a generation is as provable as any other metered call.

Endpoints

Notes

• Every image carries signed provenance. The response includes a Sable-signed provenance manifest, embedded in the image file itself, that anyone can verify at POST /v1/images/verify. It is C2PA-shaped, not a C2PA claim, and it does not survive re-encoding.
• Metered per image. Pricing is per image, published in GET /v1/images/models, and billed on the same prepaid balance as inference.
• Vision input already works today. Sending images into a model (vision) is not this endpoint: it works through ordinary chat completions for multimodal models. This page is about image generation.
• Do not assume image generation is available on a given deployment. Read GET /v1/images/models first, and treat an empty list as not enabled.