Generate images

generateImage creates images from text or edits source images with an optional mask. Omit account to use the platform Vercel AI Gateway credential. Platform generations run in the background, and the result signal emits after generation finishes. Their model cost counts toward the organization’s monthly Platform AI allowance; customer-managed accounts don’t use this allowance. The platform default is bfl/flux-2-flex and generates at most 16 images per action. Use a provider account for larger batches. If background generation exhausts its retries, the result signal fails with the provider error and prevents dependent actions from running.

Generate an image

automations/create-campaign-image.automation.ts

import {
  automation,
  generateImage,
  markSignificant,
  onDashboardRun,
  withPrerequisites,
} from "automate.ax"

export default automation("Create campaign image", () => {
  const run = onDashboardRun({ title: "Create campaign image" })

  const result = withPrerequisites(run, () =>
    generateImage({
      aspectRatio: "16:9",
      prompt:
        "A paper-cut illustration of a solar-powered city at sunrise, no text",
    }),
  )

  markSignificant(result.images[0]!)
})

An organization member starts generation from the dashboard. The first image appears in the action output, and markSignificant keeps the completed run in the default Recent runs view. Each run uses the organization’s Platform AI allowance.

Edit images

Pass source images and an optional mask in the structured prompt. Each image or mask can be a Blob, base64 string, data URL, HTTP(S) URL, Uint8Array, or ArrayBuffer. generateImage normalizes these inputs, but the selected provider and model must support image editing and masks. sourceImageBytes and maskBytes can be byte signals returned by earlier actions. Mark the edited image so the completed run stays visible:

const edited = generateImage({
  prompt: {
    images: [sourceImageBytes],
    mask: maskBytes,
    text: "Replace the masked background with a snowy mountain range",
  },
})

markSignificant(edited.images[0]!)

Inputs

Input	Type	Required	Description
`prompt`	`string \| { images: DataContent[]; text?: string; mask?: DataContent }`	Yes	Text-to-image prompt or image editing/variation input.
`model`	`GenerateImageModelId`	No	Gateway or provider-native image model ID. Available models and the default depend on `account`.
`account`	Provider account reference	No	Customer-managed provider credentials. Omit for the platform gateway.
`n`	`number`	No	Number of images to generate. Defaults to `1`. Platform AI accepts at most `16`. Provider accounts aren’t subject to that platform limit.
`size`	`${number}x${number}`	No	Requested pixel dimensions, such as `1024x1024`. Use this or `aspectRatio` according to the model.
`aspectRatio`	`${number}:${number}`	No	Requested aspect ratio, such as `16:9`. Use this or `size` according to the model.
`seed`	`number`	No	Provider-supported deterministic seed.
`maxImagesPerCall`	`number`	No	Maximum images in each provider call. `generateImage` combines split calls into one result.
`maxRetries`	`number`	No	Maximum retries for each provider call. Defaults to `2`.
`providerOptions`	`Record<string, Record<string, JSONValue>>`	No	Provider-specific image options such as quality, background, output format, or moderation.
`headers`	`Record<string, string>`	No	Additional request headers for HTTP-based providers.

Generation inputs accept compatible signals. Account selection is static. Model autocomplete follows the selected provider, but arbitrary string IDs remain valid for newly released models. The inputs mirror the AI SDK generateImage surface. generateImage also accepts Blob source images and normalizes generated files to Blob values. It omits abortSignal because actions execute durably after planning, so an in-memory AbortSignal can’t control the later provider call.

Output

images contains every generated image as a Blob. The first generated image is images[0]. Read its media type from blob.type, and use standard Blob methods such as arrayBuffer(), bytes(), or text() when you need the contents. generateImage also returns provider warnings, combined token usage when reported, providerMetadata, and a responses entry for each underlying provider call. Each response includes its start timestamp, model ID, and optional headers.

Provider accounts

Connect credentials with bunx automate.ax accounts, then pass a named account reference. API keys don’t appear in automation code or signals.

Account	Default model
Platform or Vercel AI Gateway	`bfl/flux-2-flex`
DeepInfra	`black-forest-labs/FLUX-1-schnell`
Fireworks AI	`accounts/fireworks/models/flux-1-schnell-fp8`
Google Generative AI	`imagen-4.0-fast-generate-001`
OpenAI	`gpt-image-1-mini`
Together AI	`black-forest-labs/FLUX.1-schnell-Free`
xAI	`grok-imagine-image`

Pass a named provider account and mark the generated file:

import { generateImage, markSignificant } from "automate.ax"
import { openaiAccount } from "automate.ax/openai"

const productShot = generateImage({
  account: openaiAccount("creative"),
  model: "gpt-image-2",
  prompt: "A studio product photo of a cobalt blue travel mug",
  providerOptions: {
    openai: {
      background: "transparent",
      quality: "high",
    },
  },
})

markSignificant(productShot.images[0]!)

This uses the connected OpenAI account instead of the Platform AI allowance. The provider can charge that account for each generation.

Built-in triggers

Built-in actions

Runtime utilities

Integrations

Generate an image

Edit images

Inputs

Output

Provider accounts

​Generate an image

​Edit images

​Inputs

​Output

​Provider accounts

Generate an image

Edit images

Inputs

Output

Provider accounts