Image Generator
Add Node → Generate → Image Generator
This is the node you’ll use the most. It makes still images — either from a text prompt on its own, or from a prompt plus reference images.
It arrives already wired to an empty image slot, ready to fill once you generate.
What’s on the node
| Area | What goes in it |
|---|---|
| Prompt box | Where you describe the image. Type into it directly, or wire something into the prompt handle and it turns read-only, showing the upstream text live. |
| Model and settings | The model picker, and under it whatever that model exposes — resolution, aspect ratio, quality. The controls change with the model, and switching resets them. |
| P | One only. Takes a Prompt Generator, a Text Area, a Prompt Split, or a finished result’s own prompt handle. |
| I | Numbered 1, 2, 3… for reference images, in the order they’re sent to the model. Each takes an image, a Media Group or a Media Placeholder. Fill one and a fresh empty I appears below for the next. |
A ring is hollow until something is connected and fills with colour once it is, so you can see at a glance what the node still needs. There’s more on all of this in How nodes connect.
Text-only, or with references?
Type a prompt and generate, and you get a plain text-to-image result.
Connect one or more images instead, and the node switches to image-to-image: now the model looks at your references while it generates, instead of working from words alone. This is the best way to keep something consistent across generations — a character, a product, a location — rather than hoping a text description gets it right.
Once you’ve connected references, you can call them out directly in the prompt, something like “combine the product from image 2 into image 1.”
Which model should you use?
| Type | Model | Resolution | Input | Price range | What it’s good at | |
|---|---|---|---|---|---|---|
| Pro | image-to-image text-to-image |
Nano Banana Pro | 1K · 2K · 4K | 10 | $0.14–0.24 | Google’s most capable, and the best here at plain-language prompting with references. Reach for it when consistency matters. |
| Pro | image-to-image text-to-image |
GPT Image 2 | 1K · 2K · 4K | 10 | $0.02–0.73 | The node’s default. A solid all-rounder for both text-to-image and image-to-image. |
| Pro | image-to-image text-to-image |
Nano Banana 2 | 1K · 2K · 4K | 10 | $0.045–0.14 | Same family, a step down. Still excellent with references and simple prompts. |
| Good | image-to-image text-to-image |
Nano Banana | HD | 10 | $0.038 | The original Google model. Solid for straightforward edits, and cheap enough to run test after test at good quality. |
| Good | image-to-image text-to-image |
Seedream 5.0 Pro | 1K · 2K | 10 | $0.045–0.09 | Strong image-to-image, for more control over composition and style. |
| Creative | image-to-image text-to-image |
Flux 2.0 | HD | 3 | $0.07 | Fewer reference slots, more character in the result. For a look, not a faithful copy. |
| Creative | image-to-image text-to-image |
Z-Image | HD | 1 | $0.005 | One reference, and the cheapest run here. Good for quick variations. |
| Creative | text-to-image | Ideogram 4.0 | 1K · 2K | 0 | $0.025–0.20 | Text-only, and the best choice when you need clean, accurate text inside the image. |
| Creative | text-to-image | ERNIE | HD | 0 | $0.03 | Text-only. Worth trying for a different feel from a prompt alone. |
Prices are indicative and set by the providers, who revise them: see wavespeed.ai or runware.ai for current rates.
Resolution lists every size a model can output; HD means it has no resolution setting at all and always returns roughly 1024×1024. Prices are per image, charged straight to whichever provider you’ve connected. The figures here are WaveSpeed’s rates; Runware prices its own catalogue. Where there’s a range, the low end is the smallest size at the lowest quality and the high end is 4K at the highest, so the setting you pick matters as much as the model you pick.
More advanced models can take quite a few reference images at once, but the count matters far less than the choice. References pulling in different directions give the model contradictory instructions, and you get muddle instead of the result you were after — so add an image because it says something the others don’t, not to fill the slots.
Switching to a different model resets your settings, so nothing carries over by mistake. And if you connect more reference images than a model can use, the extra ones just get greyed out rather than removed — switch to a model that supports more, and they come back to life.