Media

media.image.generate · Generate images

Generates 1 to 4 images from a brief using the brand Kit colors, the approved primary logo (useLogo) and the brand voice; referenceAssetIds must be approved Library images. Fails with budget_exceeded (detail images_per_month) when the plan monthly image allowance is used up, or budget_exceeded when the AI budget does not fit; nothing is generated then. Once the person confirms, it waits for the images and returns status completed; results land in the Library as proposed files until a person approves them. Reference images are sent to the model as images. Each asset carries an art-direction review (palette fit on the pixels, on-brand and reference match from a vision model, issue codes, verdict ok/review/redo): mention a redo or review to the person and offer another try instead of presenting it as finished.

Consumes the brand AI budget.

Input

{
  brief: string  // ≤ 2000
  format?: "square" | "portrait" | "landscape" | "story"  // default "square"
  count?: integer  // 1–4 · default 2
  useLogo?: boolean  // default false
  referenceAssetIds?: string[]  // ≤ 3 items · default []
}

Output

{
  taskId: string
  status: "working" | "completed" | "failed"
  assets: Array<{
    assetId: string  // uuid
    url: string  // uri
    review?: {
      verdict: "ok" | "review" | "redo"  // redo = a visible defect (garbled text, a drawn logo, a watermark) or clearly off brand; offer to generate again.
      paletteFit: integer | null  // Share of coloured pixels on the Kit palette.
      onBrand: integer | null
      matchesReferences: integer | null
      issues: string[]
      note: string
    }
    | null
  }>  // default []
  reservedUsd: number
}