Sample Request

curl --location 'https://cloud.leonardo.ai/api/rest/v2/generations' \
     --header 'authorization: Bearer {api-key}' \
     --header 'Content-Type: application/json' \
     --data '{
       "model": "remove-bg",
       "public": false,
       "parameters": {
         "size": "auto",
         "type": "auto",
         "format": "png",
         "guidances": {
           "image_reference": [
             {
               "image": {
                 "id": "YOUR_IMAGE_ID",
                 "type": "UPLOADED"
               },
               "strength": "MID"
             }
           ]
         }
       }
     }'

API Request Endpoint, Headers, Parameters

Endpoint

https://cloud.leonardo.ai/api/rest/v2/generations

Headers

--header "accept: application/json" \
--header "authorization: Bearer <YOUR_API_KEY>" \
--header "content-type: application/json"

Body Parameters

Parameter	Type	Definition
model	`string`	Required. Set to `remove-bg`.
parameters.bg_color	`string`	Optional. Adds a solid background colour to the cutout, as a hex code or colour name. The output is no longer transparent — pair with the `jpg` format when possible.
parameters.channels	`string`	Optional. Set to `rgba` to return the cutout image, or `alpha` to return only the greyscale alpha mask. Defaults to `rgba`.
parameters.crop	`boolean`	Optional. Set to `true` to crop the output image to the bounding box of the subject. Defaults to `false`.
parameters.crop_margin	`string`	Optional. Margin to add around the cropped subject, as an absolute value (`30px`) or relative to the subject size (`10%`). Accepts one, two, or four values. Only applies when `crop` is enabled.
parameters.format	`string`	Optional. Output image format. Set to `png`, `jpg`, or `webp`. PNG and WebP preserve transparency, JPG does not. PNG output is capped at 10 megapixels, so pair larger `size` values with `webp` to keep transparency. Defaults to `png`.
parameters.guidances.image_reference	`array`	Required. Array of exactly 1 image to remove the background from. Each item specifies an `image.id`, an `image.type` (`INIT`, `GENERATION`, `UPLOADED`, `GENERATED`, or `VARIATION`), and a `strength` (`LOW`, `MID`, or `HIGH`). `strength` defaults to `MID`.
parameters.position	`string`	Optional. Position of the subject within the output canvas, as `original`, `center`, a single percentage, or two percentages for horizontal and vertical placement.
parameters.priority	`integer`	Optional. Priority level for job queue routing.
parameters.quantity	`integer`	Optional. Number of outputs to generate. Fixed at `1`. Defaults to `1`.
parameters.roi	`string`	Optional. Region of interest to search for the subject, given as two x/y coordinate pairs in pixels (`0px 0px 100px 100px`) or percentages (`0% 0% 100% 100%`). Anything outside this rectangle is treated as background.
parameters.scale	`string`	Optional. Scale of the subject relative to the output canvas, as `original` or a percentage between `10%` and `100%`.
parameters.semitransparency	`boolean`	Optional. Preserves semi-transparent regions such as glass, smoke, and veils. Defaults to `true`.
parameters.shadow_opacity	`integer`	Optional. Opacity of the artificial shadow, from `0` to `100`. Only applies when a shadow is enabled.
parameters.shadow_type	`string`	Optional. Style of artificial shadow to render under the subject. Set to `none`, `drop`, `3d`, or `car`. Defaults to `none`.
parameters.size	`string`	Optional. Maximum output image resolution. Set to `auto`, `preview`, `full`, or `50MP`. PNG output is capped at 10 megapixels regardless of this setting. Defaults to `auto`.
parameters.type	`string`	Optional. Foreground type hint to improve cutout quality. Set to `auto`, `person`, `product`, `car`, `animal`, `graphic`, `transportation`, or `other`. Defaults to `auto`.
parameters.type_level	`string`	Optional. How specifically the subject should be classified. Set to `none` to skip classification, `1` for a coarse category, `2` for a specific category, or `latest` for the most detailed classification available.
public	`boolean`	Optional. Set to `false` to keep result private.