Introduction
Utility nodes expand your creative control within Workflows by allowing you to edit, extract, and manipulate media without leaving the Workflow Editor. These nodes bridge the gap between generation and post-production, enabling you to refine your outputs and build more sophisticated pipelines.
Use utility nodes when you need to:
- Isolate or swap audio tracks from generated videos
- Combine multiple video generations into a single sequence
- Extract specific frames for use as images or generation inputs
- Adjust video length to meet exact specifications
Utility nodes work with your generated or uploaded media, allowing you to transform videos and audio directly within your Workflow.
Extract Audio
The Extract Audio node pulls the audio track from a video input, outputting it as a standalone audio file.
Inputs: Video
Outputs: Audio
When to use it: Pull the audio track from a generated video to reuse across multiple scenes, or isolate specific sound effects or dialogue to repurpose in other projects.
How to use:
- Link a Video output to the Extract Audio node's video input
- Run the node
- The extracted audio will be available as the node's audio output
Add Audio
The Add Audio node combines a video input with an audio input, replacing the original audio track.
Inputs: Video, Audio
Outputs: Video
When to use it: Replace generated audio with custom music, voiceover, or sound design to match your creative vision or brand requirements.
How to use:
- Link a Video output to the Add Audio node's video input
- Link an Audio output to the Add Audio node's audio input
- Run the node
- The output video will contain your original video with the new audio track
Stitch
The Stitch node combines multiple video inputs into a single sequence, playing one after another in order from Input 1 to the final input.
Inputs: Video (multiple)
Outputs: Video
When to use it: Create multi-scene narratives by combining separate video generations into one continuous piece, or build product demos by stitching together different angles and features.
How to use:
- Link Video outputs to the Stitch node's inputs in your desired order (Input 1 plays first, Input 2 plays second, etc.)
- Add additional inputs as needed by clicking the + button on the node
- Run the node
- The output will be a single video with all inputs combined in sequence
Extract Frame
The Extract Frame node allows you to select and extract a specific frame from a video as a high-quality image.
Inputs: Video
Outputs: Image
When to use it: Capture a specific hero moment from your video to use as a thumbnail, marketing asset, or starting image for another generation.
How to use:
- Link a Video output to the Extract Frame node's video input
- Run the node
- Scrub through the video preview to find your desired frame
- The selected frame will be available as an image output
First Frame
The First Frame node automatically extracts the opening frame from a video input as a high-quality image.
Inputs: Video
Outputs: Image
When to use it: Grab the opening frame to verify your first frame reference was applied correctly, or use it as a consistent starting point when creating variations.
How to use:
- Link a Video output to the First Frame node's video input
- Run the node
- The first frame will automatically be extracted as an image output
Last Frame
The Last Frame node automatically extracts the final frame from a video input as a high-quality image.
Inputs: Video
Outputs: Image
When to use it: Use the final frame of one generation as the first frame input for the next to create seamless scene-to-scene continuity, perfect for building longer narratives.
How to use:
- Link a Video output to the Last Frame node's video input
- Run the node
- The last frame will automatically be extracted as an image output
Trim Video
The Trim Video node shortens a video input to a specified duration.
Inputs: Video
Outputs: Video
When to use it: Cut longer generations down to exact durations for social media specs, or remove unwanted sections from the beginning or end while keeping your core content.
How to use:
- Link a Video output to the Trim Video node's video input
- Set your desired duration (must be shorter than the original video)
- Run the node
- The output will be your trimmed video at the new duration
Reverse Video
The Reverse Video node automatically reverses both the video and audio of your input, playing it backwards from end to start.
Inputs: Video
Outputs: Video
When to use it: Create rewind effects, time-reversal sequences, or reveal a process in reverse order. Useful for creative transitions, "going back in time" narratives, or showing cause-and-effect relationships from an unexpected angle.
How to use:
- Link a Video output to the Reverse Video node's video input
- Run the node
- The reversed video (with reversed audio) will be output and ready to use
JSON Parse Node
The JSON Parse node extracts multiple values from structured JSON outputs using JSON Path syntax, allowing you to break down a single LLM response into many separate text values.
Inputs: Text (JSON)
Outputs: Text (up to 12 outputs)
When to use it: Break down a structured LLM response containing multiple scenes into individual prompts for different nodes that receive text inputs, extract different variations (time of day, weather, styles) from a single description to route to different outputs, or isolate specific text values like character descriptions, camera angles, and environments to use throughout your workflow.
How to use:
The JSON Parse node works best when paired with an LLM node that outputs structured JSON. Start by giving your LLM node a system prompt that defines the JSON format you want, then use JSON Parse to extract the individual values.
Step 1: Define your JSON structure in the LLM node
In your LLM node's system prompt, describe the task and specify the exact JSON format you want as output. For example:
Design a scene storyboard using image prompts. Use the user input to drive the plot.
Use this JSON format:
{
scene: array<{title: string, prompt: string}>
}
Step 2: Extract values with JSON Parse
Connect the LLM node's output to the JSON Parse node, then enter paths to extract the specific values you need:
-
scene.0.prompt→ extracts the first scene's prompt -
scene.1.prompt→ extracts the second scene's prompt -
scene.2.prompt→ extracts the third scene's prompt -
scene.0.title→ extracts the first scene's title
Step 3: Connect and run
Connect each extracted output to other nodes in your workflow, for example, routing each scene prompt to its own image or video generation node. Hover over each output icon to preview the extracted value.
Customizing to your needs
Both the system prompt and the JSON paths are fully customizable. You can design any JSON structure that fits your workflow. For example, you could ask the LLM to generate character descriptions, camera angles, environment details, or style variations, whatever your workflow requires. Just make sure the paths in the JSON Parse node match the structure you defined in your system prompt.
Path syntax basics
Use dot notation to navigate your JSON structure. For arrays, use the item's index number (starting from 0). For example, if your JSON has shots.0.camera, that means: go into the shots array, get the first item, and return its camera value.
Segment Image
The Segment Image node extracts specific objects or elements from an input image based on a text prompt, outputting a clean image containing only the selected content.
Inputs: Image, Text
Outputs: Image
When to use it: Isolate a subject from its background to use as a reference in another generation, extract specific props or elements from a scene for compositing, or pull individual objects from a complex image to use as assets elsewhere in your workflow.
How to use:
- Link an Image output to the Segment Image node's image input
- Enter a text prompt describing what you want to extract (e.g., "the person," "the car," "the background trees")
- Run the node
- To refine your selection, click Edit Segments — this opens a popup showing your input image with the extracted elements highlighted by a purple mask
- Use the Exclude from Mask option at the bottom of the screen to click on any masked areas you want to remove from the final selection. Clicking Exclude from Mask also opens a dropdown menu with additional options
- From the dropdown, select Isolate Mask to narrow the selection down to a single specific object
- Once satisfied with your selection, click Run again to export the refined output
- The final image will contain only the selected elements
Related articles
Next steps
Now that you understand utility nodes, explore the Workflow Library to see how these nodes work in pre-built pipelines, or experiment by adding them to your existing Workflows to streamline your creative process.