Image Gen

Image Gen produces one image (or a small set of variants) from a prompt, with up to ten reference images steering subject, style, or composition. It fronts the whole image catalog — Nano Banana Pro & 2, GPT Image 2, Seedream 4.5/5, FLUX 2 & Kontext, Ideogram v3, Imagen 4, Qwen Edit, and more — behind one consistent node.

Inputs	prompt (text), image_1 … image_10 (references)
Outputs	image

Parameters

Param	Notes
model	The image model. Reference handling and edit strength vary by family — Kontext and the -edit models are strongest at faithful edits; Nano Banana and GPT Image at instruction-following composition.
prompt	The node’s own prompt. If a text node is wired in and this field has text, the local text wins — the node shows a “local prompt overrides” banner; clear the field to let the wired value drive.
reference_images	Complements the wired reference slots.
aspect_ratio	E.g. 1:1, 16:9, 9:16 — as supported by the model.
quality	Model-dependent quality/size tier.
output_format	png / jpeg / webp where the model offers a choice.
n_variants	Generate several candidates in one run.
prompt_enhancer	off / generic / cinematic.

Only parameters the selected model actually supports are shown — a control you don’t see is one that model ignores.

Working with references

References are the difference between “an image like this” and “this product / character, in a new scene”. Wire photos into image_1…image_10 and address them in the prompt (“the jacket from the reference, worn by…”). Order matters to some models — put the most important reference first. More in Reference images.

Gemini Image

A sibling node, Gemini Image, exposes Gemini’s native image controls that the generic node doesn’t: temperature, top_p, thinking_level, and resolution, alongside the usual prompt, references, aspect ratio, and variants. Reach for it when you specifically want Gemini’s rendering and want to tune it; reach for Image Gen for everything else.