agenthub · image.generate
Generate Image
Turn a text prompt into one or more images. Wire a reference image in and the same node edits instead of inventing.
Why the node earns its place
A provider-agnostic operation rather than a diffusion primitive: one call to the image capability, with the model dropdown filtered to models that can actually do it. You pick a provider and a model; nothing about the node changes when the provider does.
It bills per produced image through the pricing spine, and emits its own usage event separate from token billing — so image spend is legible on its own rather than buried in an LLM line.
How it works at run time
- 1Take the prompt from the wired port, or from a wired message object, so an LLM can write the prompt for it. No prompt is an error, not an empty image.
- 2Look for a reference image: the wired
imageport first, then a run-level uploaded reference. A list is passed through intact rather than collapsed — that is how multi-image compositing works, e.g. a base poster plus a logo. - 3Call the model with the requested resolution and count.
- 4Return the first image as a data URI, the full array, and a usage descriptor — then emit an image-usage event for billing.
What comes out
image the first image, as a data: URI images every image produced usage model, provider, image count, billing dimensions
What you wire
Inputs
- prompt *string
- imagestring
Outputs
- imagestring
- imagesarray
- usageobject
Configuration
| Key | Type | Default & options | What it does |
|---|---|---|---|
| provider * | string | choices from model-library:providers | Provider of the image model (e.g. google) |
| model * | string | choices from model-library:image_models | Image-generation model |
| resolution | string | default "1K" 1K · 2K · 4K | Requested resolution tier (billing reflects the produced size) |
| count | number | default 1 | Number of images to generate |
What usually goes wrong
Watch for this
Resolution is a request, and billing reflects what was actually produced — some models ignore the tier entirely. Raising count multiplies the cost linearly; there is no batch discount.
Behaviour & provenance