Nodes & connections (reference)

Learn how nodes communicate, how data flows through your workflow, and how to run and manage your generations in Nodes.

This guide covers the mechanics of building flows in Nodes. If you're just getting started, read “What is Nodes?” and “How do I add and connect nodes?” first.

A typical workflow: a base image and text feed the Generator, which produces the render.

Types of nodes

Before connecting anything, it helps to know what tools you have. Each node does one specific job: some bring content in, some write the prompt, some generate something new and some transform what you already have. These are the 17, in menu order:

Image (“Upload”) — Your photo, screenshot or render: the starting point. Connected to a Generator, the first image is the base.
Reference — An image with its declared role — material, furniture, element, scene, style or winning render — so the Generator knows what to take from it.
3D model — Opens your model (.glb, .obj, .fbx, .dae, .kmz…) in a viewer; the view you choose comes out as an image.
Text — The prompt: what you want to see. It includes templates, styles and the “Advanced options” for location, materials, vegetation and light.
Image scanner — Looks at an image and writes a complete render prompt; its second mode prepares an exploded axonometric view.
Mood extractor — Reads the atmosphere of a reference photo — scene, people, light, framing — and writes it as a prompt, without copying its architecture.
Style — Outputs the prompt of one of the 18 gallery styles.
Assistant — Turns your instruction, images and notes into a detailed prompt for the Generator.
Image Generator — Creates the render with the model you choose and keeps the result history inside the node.
Video — Animates an image: the connected one is the opening frame and the text describes the movement. The clip is worked on in its panel.
List — Groups several texts or several images so a Generator runs once for each.
Enhance render — Takes a render to a hyper-realistic photo without changing the design.
Lighting — Relights the same render at another time of day.
Camera angles — Photographs the same space from another viewpoint.
Image Upscaler — Raises the resolution 2x or 4x and adds detail.
Group — A frame to organize nodes: whatever you drop inside moves with it. Deleting the group doesn't delete its nodes.
Sticky note — A text note on the canvas; it doesn't take part in running the flow.

How nodes communicate

Every node has ports — the small circles on its edges. Output ports send data out and input ports receive it. When you draw a line from an output to an input, you create a connection that carries data from one step to the next. For example, the image a Generator produces can flow into a Lighting node to change its light.

Every node has ports: the inputs, on the left, receive data; the output, on the right, sends it. The color marks the type.

Some inputs are required: the node won't run without them.
By default, each input accepts one connection. If you connect a new one, it replaces the old one.
Some inputs accept several connections: the Generator's image input (the first is the base and the rest are references), the Assistant's inputs and the List's inputs.
Outputs have no limit: you can send one result to as many nodes as you want. On nodes that generate, you can drag the output by hand once there's a result.

Data types

Data flows between nodes in two types, each with its own color. You can only connect ports of the same type — image to image, text to text: if you try to connect incompatible types, the line won't snap. There's no video type: the clip stays in its node.

Type	Color	What it carries
Image	Blue	Photos, renders and the view of a 3D model
Text	Violet	Prompts: the ones you write and the ones produced by the Assistant, the Scanner, the Mood extractor or the Style node

Wires carry their type's color: blue for image, violet for text.

Connecting nodes

There are several ways to wire nodes together. Use whichever feels fastest.

Drag from port to port: The most direct way. Click and hold an output port, drag the line across the canvas, and drop it on a compatible input port.
Drag to empty space: Not sure which node you need? Drag from any port into empty space and the node menu opens, filtered to show only nodes compatible with that port.
Move a node close, or use quick connect: If you drop a node near another one with a free compatible port, they connect on their own. And “Quick connect”, on a node's toolbar, opens the menu with the compatible nodes: the one you pick comes out already connected.

Good to know

Connections always go from output to input, never the reverse.
You can't connect a node to itself or create loops: the workflow always moves forward.
While you drag a connection, compatible ports light up in their color and the rest dim.
To remove a connection, hover over the line and click the scissors, or select it and press Delete. With the Cut tool (C) you cut several in one stroke.

What you can connect to what

Each node type accepts specific inputs and produces specific outputs. Use this as a quick reference when building your workflows.

Node	Accepts as input	Produces as output
Image (“Upload”)	— (source node)	Image
Reference	— (source node)	Image
3D model	— (the model comes in as a file)	Image (the chosen view)
Text	— (source node)	Text
Image scanner	Image (required)	Text
Mood extractor	Image (optional: also a photo inside the node, or the mood written by hand)	Text
Style	— (source node)	Text
Assistant	Images and texts (several)	Text
Image Generator	Images (several; the first is the base) and Text. It needs at least one of the two	Image
Video	Opening-frame image (required), closing-frame image (optional, on models that support it) and Text (optional)	— (the clip stays in the node)
List	Texts or images (several; one type per list)	The same type
Enhance render	Image (required) and Text with notes (optional)	Image
Lighting	Image (required)	Image
Camera angles	Image (required)	Image
Image Upscaler	Image (required)	Image

Group and Sticky note have no ports. One output can feed several inputs: for example, a Text node can go to a Generator and an Assistant at the same time.

Running your workflow

Only nodes that generate something have a run button: ▶ “Run” (↻ “Regenerate” once there's a result) runs that node. With the node selected, the ⌄ arrow next to ▶ opens “Run modes”, and each mode shows its token cost before running. The Upscaler and Video only have their button.

Mode	What it does	When to use it
Run node	Runs only the selected node, using whatever data is already on its inputs.	You tweaked one setting — like the model — and want to re-run just that step.
Run from here	Runs the selected node and everything after it.	You changed something mid-workflow and want to reprocess from there without spending tokens on earlier steps.
Ctrl + Enter	With a Generator or an Assistant selected, runs the whole connected flow from the start.	You want to run the full chain in one go.

To cancel, press “Cancel” (✕) on the node that's generating: the rest of the chain stops too, and what was already done stays in the history. If the job was still queued, its tokens are returned; if it had already started, it finishes and is delivered.

Workflow states

Each node shows its state so you can track progress at a glance.

State	What it means
Idle	Ready to run, no result yet. It has no label.
Queued	Waiting for upstream nodes to finish.
Generating	The AI is creating the result; the label shows the time left.
Completed	Done: the result appears on the node. It has no label either.
Error	Something went wrong: hover over the “Error” label to see why.
Cancelled	You cancelled the run.

Batch processing with List

Connect a List with several items to a Generator and it runs once for each checked item. In a text List, each item is added to the prompt; in an image List, each image goes in as a reference over the same base (or becomes the base, if there's no other). Each item is a render and is charged, and variants don't apply. For now, batches work with NanoBanana PRO and NanoBanana Turbo.

Tokens and cost

Nodes that generate use tokens, and you see the cost before running: on the Generator's model chip, in the run modes menu, in the ⚙ panel and on the label of Enhance render, the Upscaler, Video and Camera angles. On the Generator, the model, the resolution, the variants and the items of a List all count; GPT Image and Seedream 5 also add 5 tokens for each extra image you connect. Lighting costs 20, Camera angles 10, and video is charged per second. The Assistant, the Image scanner, the Mood extractor, the Style node and the Agent don't charge: the node that generates afterwards does. The price of each model is in “AI models”.

Tips and best practices

Discover compatible nodes from a port. Drag into empty space and the menu opens filtered by data type.
Use “Run from here” when iterating: it saves time and tokens because it doesn't repeat the earlier steps that didn't change.
Branch your workflow. Connect one output to several inputs to process the same data in different ways and compare.
Follow the colors. Blue is image and violet is text: they help you trace the flow at a glance.
On the Generator, the first connected image is the base and the rest are references; each one's type is chosen on its node or in the ⚙ panel. To change the base, reconnect the images in another order.