Image Generation Node

The Image Generation node creates a new image from a text prompt and an optional reference image. It is provider-bound: each node carries a provider preset that routes the request to one of the two image backends, and the generated result is stored on the canvas as a synced media asset.

What it does

• Resolves a prompt from the node field or a connected text input.
• Resolves the provider and model from the node's preset, then posts a generation request to the matching API route.
• Resolves a single connected reference image when the selected model supports image input.
• Decodes the returned image, stores it durably, and emits it on the image-out port.

Providers

Image generation is wired to two providers. The node preset selects which one a given node uses.

fal models

These fal image endpoints have first-class request contracts. Endpoints not in the contract list are still accepted and sent to fal as-is, using an image_size sizing mode and an image_url reference field by default.

Inputs

• Prompt — required. Read from the node prompt field or a connected text source. Prompts must be under the shared maximum prompt length.
• Reference image — optional. The first connected visual input is resolved and passed to the model when the model supports image input. fal endpoints without a reference image field reject reference input with a 400.
• Aspect ratio and image size — node fields, shaped per provider. fal uses either an image_size preset or an aspect_ratio string depending on the endpoint contract.
• Model parameters — endpoint-specific extras such as guidance_scale, num_inference_steps, and seed. Unsupported keys for the chosen endpoint are filtered out before the request is sent.

Outputs

On success the node emits the generated image on the image-out port and stores it as durable media so downstream nodes and later sessions can reuse it. If the prompt looks like JSON, HTML, or source code rather than a visual description, the node attaches a prompt-quality warning to the result instead of failing.

Run a generation

1. 1

   Add the node and set a prompt

   Add an AI Image node, then type a prompt or connect a text source. The node opens its configuration panel automatically so you can pick a model and sizing.

2. 2

   Choose provider, model, and size

   The provider preset determines the backend route. Select an aspect ratio or image size and any supported model parameters. Connect a reference image if the model accepts one.

3. 3

   Run the node

   Google requests return inline. fal requests are submitted to the fal queue and polled to completion, which tolerates client disconnects during longer jobs.

4. 4

   Reuse the output

   The stored image flows out of image-out. Wire it into a Video Generation node, an audio workflow, or an Image node for cropping and export.

Async and polling (fal)

The fal image route defaults to the asynchronous queue path. The Builder Studio executor submits the job, receives a request id and a signed status token, then suspends and polls the fal status endpoint until the image is ready. Google image generation runs inline within the request.

Agent and API notes

Agents drive image generation by setting the node prompt and provider preset, then running the node. The example below is a direct fal route body. Reserved internal keys are stripped, unknown model parameters for the chosen endpoint are dropped, and the request is bound to a canvas for credential and access scoping.

1{2  "prompt": "a calm desert at golden hour, cinematic",3  "model": "fal-ai/flux-2",4  "imageSize": "landscape_16_9",5  "modelParams": {6    "guidance_scale": 3.5,7    "num_inference_steps": 28,8    "seed": 429  }10}