> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tryblend.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Utility
> Run a deterministic helper operation — extract audio from a video or compose text from a template.
The **Utility** node runs a fixed, deterministic helper: no AI model, no randomness.
You pick one operation from a dropdown, and the node's handles and settings reshape to match it. Same inputs always produce the same output, which makes Utility nodes the reliable glue between model-driven steps.
Like other processing nodes, a Utility runs once per input item group. One incoming video extracts one audio file; three batched videos make Extract Audio run three times.
If Compose Text receives lists on multiple variable handles, the combinations multiply.
## Operations
Pick the operation from the **Utility** dropdown at the top of the node's settings panel. The two available operations are:
Pulls the audio track out of a video and outputs it as a standalone audio file.
* **Input:** one video — upload it directly in the panel, or connect a video output to the **Video** handle.
* **Output:** one MP3 audio file on the **Audio** handle.
* **Settings:** audio **format** (MP3) and **bitrate** in kbps (32–320, default 96).
* Videos with no audio stream fail with a clear error.
Combines connected text inputs into a single text output using a reusable template.
* Write a template and insert variables like `{{input_1}}`. Each variable you add creates its own text input handle on the left.
* Use **Add input** to append the next variable (`input_1`, `input_2`, …) to the template automatically.
* At run time, each variable is replaced by the text connected to its handle. The default template is `{{input_1}}\n{{input_2}}`.
* **Output:** the composed string on the **Text** handle.
## Inputs
Inputs depend on the selected operation.
**Extract Audio**
| Handle | Type | Required | Notes |
| ------ | ------- | -------- | -------------------------------------------------------------- |
| Video | `video` | Yes | One video file. Upload in the panel or connect a video output. |
**Compose Text**
| Handle | Type | Required | Notes |
| ----------------------- | ------ | -------- | --------------------------------------------------------------------------------------------- |
| `input_1`, `input_2`, … | `text` | No | One handle per `{{variable}}` in the template. Add or remove variables to change the handles. |
## Outputs
| Operation | Handle | Type | Notes |
| ------------- | ------ | ------- | --------------------------------------------- |
| Extract Audio | Audio | `audio` | One MP3 file at the configured bitrate. |
| Compose Text | Text | `text` | The template with every variable substituted. |
## Settings
| Operation | Setting | Notes |
| ------------- | ------------ | ------------------------------------------------------------------------------------------- |
| Both | Utility | Choose the operation. Switching it resets the node's settings and handles. |
| Extract Audio | Audio format | MP3 — keeps the file small and broadly compatible. |
| Extract Audio | Bitrate | Target audio bitrate in kbps. Allowed range 32–320 (default 96). |
| Compose Text | Template | The text to output, with `{{input_1}}`-style variables. **Add input** appends the next one. |
## When to use
* **Extract Audio** — feed a video's soundtrack into a transcription or audio model, or output the audio on its own.
* **Compose Text** — merge several text streams (prompts, model outputs, captions) into one block with fixed wording around them, without spending a model call.
## Connects to
* **Extract Audio**: upstream a video source (an Input or model node); downstream a transcription model, audio model, or User Result node.
* **Compose Text**: upstream any text-producing nodes into each variable handle; downstream a model prompt or a User Result node.
For the exact fan-out rules, see [Lists & single values](/concepts/lists-and-values).
**Extract Audio** runs on the server and supports the node's standalone Run action. **Compose Text** is resolved locally as part of the graph and does not support a single-node run.
## Next steps
How nodes, handles, and connections fit together.
Capture a utility's output as a workflow result.