Prompt bar

When you select a node that can generate, a “prompt bar” appears below it. Write the prompt, pick a model, adjust settings, check the estimate and click send, all from there.

What's in the prompt bar

The prompt bar has three rows, from top to bottom:

• Top row: the “@” button (opens the element picker), the add-image button, and thumbnails of the reference assets connected to the node (Image 1, Image 2, …). Video nodes also have an add-video button, and an attachment button appears when the model accepts documents. In first-and-last-frame mode, the add-image button becomes two fixed slots, “First frame” and “Last frame”, which you can swap;
• In the middle is the prompt box: describe the image, video or speech you want. Type @ to reference a connected node, an element, or an asset from your own assets (see the “References and @-mentions” chapter);
• On the left of the bottom row is the model picker. Models differ in what they can do and what they cost; you always see a model's display name. If you haven't picked one, the default model for that task type is used;
• To the right of the model is a settings summary (for example “Image generation · 1K” or “16:9 · 1080P · 5s”). Open it to adjust the generation mode, aspect ratio, resolution, quality, duration, output format and so on; only the settings the model supports are listed. On voice nodes this area holds the voice and “Adjust”, and some models also have “Style”. “Camera” appears only when the node has camera settings turned on. On narrow screens, less-used buttons move into “More”;
• On the right of the bottom row are “Prompt optimization” (shown only when the selected model has it set up; you can restore the original after optimizing), expand / collapse, the image count, and the credit estimate with the send button. The estimate updates as you make changes; click send to submit.

Estimates and billing

Prices and pricing units depend on the selected model. Usage-based tasks are settled by actual usage, and any unused precharged credits are returned.

• The estimate shows in one of four ways: “—” (no estimate right now), “N”, “≈ N” and “Up to N”, with a credit icon beside the number. Usage-based models (billed per token, per minute, per 5 seconds and so on) precharge first. When the expected usage can be worked out, you see “≈ N”; hover over it to see how much is precharged, that it's settled by actual usage when the task finishes, and that anything unused is returned. When no expected value can be worked out, you see “Up to N”. The settlement after the task finishes is what counts;
• When a model has no price yet or its pricing is misconfigured, the estimate shows a matching notice.

Submitting and results

After you click send, the node shows “Queued”, then “Generating”, and the result comes back to the node when it's done.

• Canceling: you can cancel while the request hasn't been sent to the model yet, and the precharged credits are returned. Once the request has been sent, it can't be withdrawn through the normal cancel flow.
• A failed task isn't always fully refunded. Usage that's already confirmed may still be charged, and when the result is unknown, the precharge is held until it's been checked. “Charged N credits, returned M credits to your balance” on a node means the task was settled by actual usage and the unused precharge was returned. “Refunded N credits to your balance” means a refund. While the refund amount is still being checked, you'll see “Refunded. The amount is pending review, so it isn't shown yet”. All of these are credits, not cash refunds.
• Double-clicking send doesn't charge you twice.