Image Generation

OpenDunes generates images through the dedicated POST /v1/images/generations endpoint, using models built for image output. Describe what you want in a text prompt and receive the image in the response.

Endpoint#

POST https://opendunes.com/v1/images/generations

The request and response are compatible with the OpenAI images API, so any SDK targeting that spec works unchanged.

Request body#

curl https://opendunes.com/v1/images/generations \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-2.5-flash-image",
    "prompt": "A terracotta geometric gate in a Saharan dune landscape, golden hour, architectural illustration.",
    "n": 1,
    "size": "1024x1024"
  }'

Response#

Each image comes back in the response itself: its bytes base64-encoded in b64_json, its type in media_type. OpenDunes also keeps a copy in Files and puts its id in file_id beside each image, so you can fetch it again later with GET /v1/files/{id}/content. A kept image never expires; it stays until you delete it.

{
  "created": 1750000000,
  "data": [
    {
      "b64_json": "iVBORw0KGgoAAAANSUhEUgAA...",
      "media_type": "image/png",
      "file_id": "019383ac-5678-7000-b000-000000000002"
    }
  ]
}

A key with storage switched off keeps nothing it gets back, so its images carry no file_id.

If the provider returns no image, the request fails with 502 no_image and you are not charged.

Pricing#

Image generation is billed in DA per image — n images cost n × the model's per-image rate. Rates vary by model; see the pricing reference and the model detail page.

Chat image output#

Some chat models can also return images inline by setting "modalities": ["image", "text"] on a POST /v1/chat/completions request — the image arrives as an images[] part on the assistant message. That is a chat capability, distinct from this dedicated image surface.

Finding image generation models#

GET https://opendunes.com/api/models?modality=image

Browse results at /models filtered to the image surface.