Image generation

Generate images from a text prompt, or edit an existing image with a prompt, using any image model on the network. The endpoints mirror OpenAI's image API.

POST /v1/images/generations   (JSON)
POST /v1/images/edits         (multipart)

Requests are routed only to services a provider declared as type: image. inference.club stores every generated image in object storage and, by default, returns a URL you can drop straight into an <img> tag.

Generations

application/json:

Field	Required	Description
`prompt`	yes	The text description of the image.
`model`	yes	An image model id from `GET /v1/models`.
`n`	no	How many images to generate (clamped server-side).
`size`	no	e.g. `1024x1024`. Passed through to the provider.
`response_format`	no	`url` (default) or `b64_json`. See below.

curl

curl https://api.inference.club/v1/images/generations \
  -H "Authorization: Bearer $INFERENCE_CLUB_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "my-image-model", "prompt": "a watercolor fox", "size": "1024x1024" }'

Python (openai SDK)

from openai import OpenAI

client = OpenAI(base_url="https://api.inference.club/v1", api_key="<your-api-key>")
img = client.images.generate(model="my-image-model", prompt="a watercolor fox")
print(img.data[0].url)

Edits

multipart/form-data — supply a source image (and optional mask) plus a prompt:

Field	Required	Description
`image`	yes	The source image to edit (png/jpeg/webp, up to 25 MB).
`prompt`	yes	How to change it.
`model`	yes	An image model id.
`mask`	no	Optional transparency mask.
`n`, `size`, `response_format`	no	As above.

curl https://api.inference.club/v1/images/edits \
  -H "Authorization: Bearer $INFERENCE_CLUB_API_KEY" \
  -F model=my-image-model \
  -F image=@photo.png \
  -F prompt="make the sky a sunset"

Response

{
  "created": 1780266332,
  "data": [ { "url": "https://api.inference.club/api/inference/assets/42/" } ]
}

url (default): the image is stored in inference.club object storage and the response returns its URL. Generated images are public by URL (so they embed in <img> tags and can be shared) — request-level privacy controls are coming.
b64_json: set response_format: b64_json to also get the raw base64 bytes inline. The image is still stored.

Every request is recorded as an inference request with the prompt, the source image (for edits), and the output image(s), visible in your dashboard. Generation is metered by image count.

Errors

`type`	When	HTTP
`missing_prompt`	No `prompt`	400
`missing_file`	`/edits` with no `image`	400
`file_too_large` / `request_too_large`	Source image or prompt over the limit	413
`unsupported_media_type`	Source image isn't png/jpeg/webp	415
`no_provider`	No online image provider serves the model for you	404
`upstream_error`	The provider's image server failed	502

Not supported (yet)

/v1/images/variations, streaming/progressive previews, and async job polling. Generation is synchronous — the request returns once the image is ready.