Image generation and editing API

/ API docs

OpenCheese serves the OpenAI Images API for every image model enabled for your account, with the same API key as text and code models.

Find the image models and their parameters

GET /v1/images/capabilities lists every image model your key can use and the parameters each one accepts, such as size, aspect ratio, image size, quality, background and output format, and how many images one request may return. Requests are checked against the same list, so send only the parameters a model names.

bash

curl https://api.opencheese.dev/v1/images/capabilities \
  -H "Authorization: Bearer $OPENCHEESE_API_KEY"

An abridged answer:

json

{
  "models": [
    {
      "publicModelId": "IMAGE_MODEL_ID",
      "displayName": "Example Image Model",
      "supportsEdit": true,
      "maxImagesPerRequest": 1,
      "params": [
        { "kind": "enum", "id": "aspect_ratio", "values": ["1:1", "16:9", "9:16"], "initial": "16:9" },
        { "kind": "enum", "id": "quality", "values": ["auto", "low", "medium", "high"], "initial": "auto" }
      ]
    }
  ]
}

Each entry in params is a parameter you may send under its id, with the values it accepts. initial is a sensible starting value; default is what applies when you leave the parameter out. maxImagesPerRequest is the largest n, and supportsEdit says whether the model takes /v1/images/edits.

Generate an image

bash

curl https://api.opencheese.dev/v1/images/generations \
  -H "Authorization: Bearer $OPENCHEESE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "IMAGE_MODEL_ID", "prompt": "A wedge of cheese on a circuit board"}'

model and prompt are required. size, quality, background, output_format, aspect_ratio, image_size and n are accepted where the model lists them.

The response carries each image as base64 in data[].b64_json; hosted image URLs are not issued, so response_format accepts only b64_json. opencheese_cost_usd in the same body is what the request cost you, in credits (USD). To save the first image:

bash

curl ... | jq -r '.data[0].b64_json' | base64 --decode > image.png

Name the file after the format you asked for in output_format.

With the OpenAI Python SDK:

python

import base64
from openai import OpenAI

client = OpenAI(base_url="https://api.opencheese.dev/v1", api_key="YOUR_OPENCHEESE_API_KEY")
result = client.images.generate(model="IMAGE_MODEL_ID", prompt="A wedge of cheese on a circuit board")
with open("image.png", "wb") as file:
    file.write(base64.b64decode(result.data[0].b64_json))

Edit images

POST /v1/images/edits takes multipart/form-data: the source image in image (or several in image[]), with model and prompt as form fields.

bash

curl https://api.opencheese.dev/v1/images/edits \
  -H "Authorization: Bearer $OPENCHEESE_API_KEY" \
  -F "model=IMAGE_MODEL_ID" \
  -F "prompt=Turn the scene into night" \
  -F "image=@input.png"

Uploads must be PNG, JPEG or WebP: up to 5 images, 10 MB each.
Other file fields, such as mask, are rejected.

Limits

Image responses are not streamed: stream and partial_images are refused.
Some models return one image per request; /v1/images/capabilities says how many n allows.
moderation cannot be changed; the model's default applies.
A prompt the model's own content policy declines is refused with moderation_blocked and not charged. A few models check the finished picture instead and charge for one they reject: that answer is moderation_blocked_billed.

Image models in chat apps

Chat apps that list every model in one picker can use an image model in a normal chat: on /v1/chat/completions and /v1/responses the latest user message becomes the prompt and the picture comes back in the reply. See Chat Completions for the reply's shape.

Pricing and usage

Image models are priced per image, by resolution and quality. The prices for your account are on the dashboard's Prices page after you sign in, and every generated image appears under Transactions with its cost.

The cheese has holes. Your infrastructure should not.