Updated October 1, 2026

Image generation

Generate images from a chat node in Image mode, use images on the canvas as references, and keep every version next to the conversation that made it.

Visual work is part of exploring an idea: concept art, thumbnails, diagrams, a mood image for a scene you're writing. In Image mode, a chat node turns your prompts into pictures, and each one lands on the canvas as its own Image node, connected to the conversation that inspired it. Nothing gets overwritten, so every version and its prompt stay in view.

Switch a chat node to Image mode

Choose Image in the mode selector at the bottom left of the composer, or press ⌘3 (Ctrl+3 on Windows and Linux) while the chat node is active. The mode is saved per node.

The model dropdown then lists image models, grouped by provider, with a "Search image models..." box. The list changes over time; a Gemini Flash image model is the default.

Generate an image

Describe what you want and send. State concrete settings when they matter, such as "16:9", "4K" or "PNG". Naming a different image model in the prompt switches the dropdown to it.

Each prompt creates one image in a new Image node beside the chat. The node shows "Generating image…" with a Cancel button, then the image with its prompt underneath. A chip in the chat tracks the status; click it to jump to the image. You can keep sending prompts while an image generates. Image node lists every state the node can show.

A text model reads your request first: the node's chat model (Auto unless you changed it). To change it, switch to Chat, pick a model and switch back; your image model stays selected. If the request is unclear, or asks for something the image model can't do, you get a question first, either as a chat reply or as Before I generate in the Image node. Click an option or type in "Or tell me what to change…". Nothing is spent on the image until you answer.

Use reference images

Connect an Image node into an Image-mode chat to edit it or use it as a reference. Uploaded, pasted and generated images all count, including ones that reach the chat through a chain or a group. A connected Scratch pad is context for the conversation, not a reference picture.

A connected image is used only when your request calls for it. If several are connected, say which one you mean: "edit the fox image", "use the second image for the style". If the model can't take reference images, Before I generate offers one that can, or Generate without them.

Iterate and branch

Follow-up prompts in the same chat make new images that build on the conversation. Each is a fresh generation, not an edit of the last picture.
Edit on a finished image creates a new Image-mode chat already connected to it, so your next prompt edits that exact picture.
Submit in new branch (⌘⇧Enter, or Ctrl+Shift+Enter on Windows and Linux) tries another direction in a branch that keeps Image mode and the image model. See Branching.

Cost and requirements

Hosted image generation needs an active plan: it's included with the trial, the yearly subscription and Lifetime. Other accounts see Upgrade to generate images when they send.

Each generation is paid from your hosted wallet at the model's price. Failed and cancelled generations aren't charged. The text model that reads your prompt is billed like any other chat message.

When the wallet runs out, the Image node shows Out of usage balance: "Your hosted wallet has insufficient funds for image generation. Add funds or use your own OpenRouter key in Settings → Providers. Your license is unchanged." That page is now Settings > API Providers. Add funds opens your account in the browser. During the trial you can't add funds yet, so the node offers Upgrade instead. See Plans and billing.

Use your own OpenRouter key

In Settings > API Providers, choose OpenRouter and save your API key.
Turn on Use my OpenRouter key for image generation.

Generations then bill your OpenRouter account, and finished images say "via your OpenRouter key". With the switch on but no key saved, the wallet pays. The switch is greyed out on plans without it, and a key alone doesn't enable Image mode.

Where generated images are saved

Each image is saved as a file in an images folder inside the canvas's folder. Click Reveal in Folder in the Image node's header to show the file, or use the copy button on the image.