image role, and appears in the tool catalog only once one is configured. Pass refs from earlier results in images to edit or combine them, with a prompt that says what to change. A value the model doesn’t support, such as an aspect_ratio or size it can’t produce, is a tool error that lists the values it does support, so an agent can correct the call and retry.
With path, each image is also written into the workspace, with the extension of the type the model returned (a JPEG asked for at site/hero.png lands at site/hero.jpg, and the result’s path says where); when count is over 1, later images get -2, -3 and so on before the extension. A path must be relative and stay inside the workspace.
Input
string
required
What to draw, or what to change in the given images.
string[]
Refs of images to edit or use as references, such as blob:sha256-….png from an earlier tool result.
string
Width to height, such as “1:1”, “16:9” or “9:16”. Omit for the model’s default.
string
Output resolution, such as “1K”, “2K” or “4K” (uppercase K). Omit for the model’s default.
integer
default:"1"
How many images to generate.
string
A path inside the workspace to also save the image to, such as site/hero.png. The extension follows the images type, so a JPEG lands at site/hero.jpg. With a count over 1, later images get -2, -3 before the extension.
Output
The images the model drew, plus any text it wrote alongside them.object[]
One entry per image, in the order they were generated.
string
The image’s file reference, such as
blob:sha256-….png. Put it in a reply as a markdown image’s URL, or pass it to another tool.string
The image’s type, such as
image/png.integer
The file’s size.
integer
Width in pixels.
integer
Height in pixels.
string
Where the image was saved in the workspace, when the call set
path.string
Text the model returned with the images, when it wrote any.