Your first image

Let's make one image end to end. This takes about a minute and costs roughly $0.03 on Gemini's cheapest image model.

#The fastest path: the starter graph

The setup wizard offers a starter graph on its last step — a ready-wired Prompt → Refine → Text→Image → Preview chain on gemini-3.1-flash-lite-image (Nano Banana 2 Lite). Click Add starter graph, then hit ▶ Run in the toolbar. Done. (Re-open the wizard any time from File ▾ → Setup wizard….)

If you'd rather build it by hand, here's the same thing in three nodes.

#Build it by hand

1. Add a Prompt node. Press Tab over the canvas (or + Node) and pick Prompt. Type a description:

   A sunlit modern villa at golden hour, floor-to-ceiling glass, a calm reflecting pool, photographic

2. Add a Text→Image node. Drag from the Prompt node's output onto empty canvas — the add menu opens filtered to compatible nodes; pick Text → Image. It wires up automatically. In the node, choose provider Gemini and model Nano Banana 2 Lite (gemini-3.1-flash-lite-image).
3. Add a Preview node the same way, wired from the Text→Image output.
4. Run. Click ▶ Run (or the ▶ on the Text→Image node to run just it). The engine calls Gemini with your key, downloads the image to your project's outputs folder, and shows it inline. The node badge reports the cost.

That's the whole loop: prompt → generate → preview.

#What just happened

• The engine made one call to Gemini on your key and saved the result to outputs/ in your active project.
• The node cached the result against a hash of its inputs. Run again and it's free — nothing changed, so nothing is re-billed. Change the prompt and only that node re-pays. See Never pay twice.
• The generation was recorded in your project ledger — click the 🕘 on the node to see its history and re-pin any past result.

#Troubleshooting

• 429 RESOURCE_EXHAUSTED on the first call → your Gemini key isn't billing-enabled (the free tier has zero image quota). Enable billing in Google AI Studio. See Google Gemini.
• A ⚠ "Add your … key" button on the node → that provider isn't configured; click it to jump to the wizard.
• finishReason: OTHER on an image edit → a silent content refusal, usually triggered by an input image (a real person's name in facade lettering, a face, a logo) rather than the prompt. Remove or crop the offending image.

#Next

• Refine your prompts with the Prompt Helper.
• Image generation — editing, inpainting, composing.
• Multi-Shot — a contact sheet of viewpoints from one scene.