Prompt Input Panel Guide

The prompt panel is where you control what the AI generates. It opens on the right side of the canvas when you select an image or video node.

Opening the Panel

Select any image or video node. The prompt panel appears on the right with three main sections:

• Prompt editor — where you write your instructions
• Model selector — choose which model to use
• Parameter controls — adjust settings like resolution, duration, and style

Rich Text Editing

The prompt editor supports structured text formatting. You can use line breaks and organization to structure complex prompts. The editor is designed for long, detailed instructions rather than short keyword-style prompts.

Using @ Mention References

This is the most powerful feature of the prompt panel. Type @ in the editor to open a dropdown listing all upstream nodes connected to this node.

Selecting a node from the dropdown inserts a reference that feeds that node's output into the current prompt. For example:

• Connect an image node to a video node
• Type @ in the video node's prompt
• Select the image node from the dropdown
• The video generation now uses that image as input

References show the node's thumbnail and name, making it easy to identify which asset you are referencing.

Reordering References

When a node has multiple upstream connections, the referenced nodes appear as cards above the prompt editor. You can drag these cards to reorder them.

The order matters: it controls how inputs are combined in the generation process. For image-to-video workflows, the first reference is typically the primary visual input.

Other Panel Controls

Below the prompt editor you will find:

• Model dropdown — select the generation model
• Style presets — apply predefined visual styles
• Parameter sliders — adjust generation parameters specific to the selected model
• Generate button — run the generation with the current settings

Tips

• Use @ mentions to build pipelines: text → image → video
• Reference ordering is especially useful when combining multiple images
• Try different models for different types of output — each model has strengths