Image: compress response images

Re-encode and optionally resize assistant output images in responses and streams.

What it does

This transform re-encodes assistant output images in the response or stream and optionally resizes them. The options match image_compress_input.

Reference

Property	Value
Type ID	`image_compress_output`
Phase	`response`
Scopes	Provider, API key

When to use

Image-generation responses are large and clients pay for transfer.
You want a uniform output format, for example WebP, regardless of the upstream format.

Configuration

Option	Type	Required	Default	Description
`output_format`	string	No	`"original"`	`original`, `jpg`, `jpegxl_lossless`, `jpegxl`, `webp_lossless`, `webp`, or `png`.
`max_edge_px`	integer	No	—	Maximum width or height in pixels. Omit to keep the original dimensions.
`jpeg_quality`	integer	No	`80`	Quality 1-100 for JPEG output.
`jpegxl_quality`	integer	No	`90`	Quality 1-99 for lossy JPEG XL output.
`jpegxl_effort`	integer	No	`7`	JPEG XL encoding effort 1-10. Higher values compress more slowly.
`webp_quality`	integer	No	`80`	Quality 1-100 for lossy WebP output.
`skip_if_smaller`	boolean	No	`true`	Keep the original image when compression does not reduce the payload size.

Example rule

{
  "transform": "image_compress_output",
  "enabled": true,
  "phase": "response",
  "models": [
    "*image*"
  ],
  "config": {
    "output_format": "webp",
    "webp_quality": 80
  }
}