Generate images from text prompts

Usage example

{
  "steps": {
    ":original": {
      "robot": "/upload/handle"
    },
    "inpainted": {
      "robot": "/image/generate",
      "use": [
        {
          "name": ":original",
          "as": "image"
        },
        {
          "name": ":original",
          "as": "mask"
        }
      ],
      "model": "google/nano-banana-pro",
      "prompt": "Replace the masked area with a breaching whale. Keep the rest of the image unchanged.",
      "format": "png"
    }
  }
}

Parameters

• interpolate

  boolean | Record<string, boolean>

• output_meta

  Record<string, boolean> | boolean | Array<string>

• queue

  batch

• force_accept

  boolean (default: false)

• ignore_errors

  boolean | Array<meta | execute> (default: [])

• use

  string | Array<string> | Array<object> | object

• model

  string

• prompt — required

  string

• format

  jpeg | jpg | png | gif | webp | svg

• seed

  string | number

• aspect_ratio

  string

• height

  string | number

• width

  string | number

• style

  string

• num_outputs

  string | number

• provider

  string (default: "auto")

Demos

Related blog posts

Inpaint an image by uploading an original image and a mask image, then use both files in /image/generate.

Controls whether Assembly Variables are interpolated for individual instruction fields.

By default, most Robot instruction fields interpolate Assembly Variables. Set this to false to treat every instruction field as literal text, or set an individual field path to false to treat only that field as literal text. For Robot-specific fields that are literal by default, set this to true or set that field path to true to opt back into interpolation.

Use field names such as path, or dotted paths such as ffmpeg.vf for nested objects.

Setting the queue to 'batch', manually downgrades the priority of jobs for this step to avoid consuming Priority job slots for jobs that don't need zero queue waiting times

Force a Robot to accept a file type it would have ignored.

By default, Robots ignore files they are not familiar with. 🤖/video/encode, for example, will happily ignore input images.

With the force_accept parameter set to true, you can force Robots to accept all files thrown at them. This will typically lead to errors and should only be used for debugging or combatting edge cases.

Ignore errors during specific phases of processing.

Setting this to ["meta"] will cause the Robot to ignore errors during metadata extraction.

Setting this to ["execute"] will cause the Robot to ignore errors during the main execution phase.

Setting this to true is equivalent to ["meta", "execute"] and will ignore errors in both phases.

Specifies which Step(s) to use as input.

For inpainting, provide both the source image and mask through use, typically with:

{
  "use": [
    { "name": ":original", "as": "image" },
    { "name": ":original", "as": "mask" }
  ]
}

Best practice:

• Tag source and mask inputs explicitly using as (or semantic upload field names)
• Keep the prompt focused on what should change in the masked/transparent region
• Leave the model/provider choice to the robot defaults unless you have a specific need

The AI model to use. Defaults to google/nano-banana. Supported models include flux-1.1-pro-ultra, flux-schnell, recraft-v3, google/nano-banana, google/nano-banana-2, google/nano-banana-pro, openai/gpt-image-2, and stability-ai/stable-diffusion-inpainting. The legacy alias gpt-image-2 is also accepted for backwards compatibility.

Prompt describing the desired image. For inpainting, describe what should appear in the masked/transparent region and that the rest should stay unchanged.

Output format. Defaults depend on model: png for Google models and openai/gpt-image-2, svg for recraft-v3, jpeg for others. Google models currently return PNG only.

Seed for the random number generator.

Requested output aspect ratio. For Google models, width/height can also be used and orientation is derived automatically when aspect_ratio is omitted.

Requested output height in pixels (mainly used by Google image models and openai/gpt-image-2).

Requested output width in pixels (mainly used by Google image models and openai/gpt-image-2).

Style of the generated image.

Number of output variants to generate (1-10).

Chooses the best provider based on your request

Allows you to specify a set of metadata that is more expensive on CPU power to calculate, and thus is disabled by default to keep your Assemblies processing fast.

For images, you can add "has_transparency": true in this object to extract if the image contains transparent parts and "dominant_colors": true to extract an array of hexadecimal color codes from the image.

For images, you can also add "blurhash": true to extract a BlurHash string — a compact representation of a placeholder for the image, useful for showing a blurred preview while the full image loads.

For videos, you can add the "colorspace: true" parameter to extract the colorspace of the output video.

For videos, you can also add "interlaced": true to detect whether the video is interlaced. This combines the cheap ffprobe field_order flag with a bounded idet sampling pass over the first frames of the source, exposing interlaced, field_order, and a diagnostic interlace_detection object under file.meta. This is computationally expensive and billed accordingly.

For audio, you can add "mean_volume": true to get a single value representing the mean average volume of the audio file.

You can also set this to false to skip metadata extraction and speed up transcoding.