<aside> 🛠

Wrap Dynamic Text in a Reusable Template

Build prompts and structured strings by combining a fixed template with a dynamic text input. Mark where the dynamic content goes with a placeholder token (default <<<input>>>); Text Decorator swaps the placeholder for the live text and outputs the finished string. Perfect for assembling LLM prompts where the instructions stay constant but the data changes - OCR text, model outputs, user queries, anything.

</aside>

What it does

Text Decorator takes a template (any text containing a placeholder token) and a dynamic text input, then produces an output where every occurrence of the placeholder is replaced with the dynamic text. The placeholder token is configurable on the node (default: <<<input>>>). This is the classic "string template" pattern - write a prompt once, reuse it across many inputs.

Problem it solves

• Reusable prompt templates – Write the instruction text once, drop it into Text Decorator, and reuse it across every input in a batch
• Clean LLM prompt assembly – Wrap OCR / model outputs / image descriptions in instruction text without manual concatenation
• Decouple instructions from data – The template lives in one node (or a Text Input); the variable content comes in through the second input — change either without touching the other
• Predictable placeholder behavior – Replacement is exact and deterministic; what you see in the template is what gets sent to the next node
• Configurable token – Use <<<input>>> (default), {{INPUT}}, [[DATA]], or any custom marker that doesn't collide with your actual content
• OCR → LLM pipelines – The canonical pattern: read text from an image (OCR), wrap it in a structured prompt (Text Decorator), then send it to an LLM
• Multi-stage workflows – Use the output of one model as the dynamic input to a templated prompt for the next model

Input/Output

<aside>

• Input: Template text + dynamic text
  • Template (top input port): The fixed text containing the placeholder token (e.g., from a Text Input node)
  • Dynamic text (bottom input port): The text to substitute in - typically the output of an OCR node, an LLM, an image-understanding model, or any other text-producing node
  • Behavior: Every occurrence of the placeholder token in the template is replaced by the dynamic text </aside>

<aside>

• Output: Single composed text string
  • Format: Plain text
  • Content: The template with every <<<input>>> (or whichever token you configured) replaced by the dynamic input </aside>

Configuration Options

• Placeholder token: The string in the template that Text Decorator searches for and replaces.
  • Default: <<<input>>>
  • Behavior: Every exact match of this string in the template is replaced with the dynamic text input
  • Custom values: Any string that doesn't naturally appear in your content (e.g., {{INPUT}}, [[DATA]], %PAYLOAD%, ___INPUT___)
  • Tips:
    • Use a distinctive token that won't show up in real text (the triple-angle-bracket default <<<input>>> is intentionally hard to collide with)
    • If you have multiple variables you want to substitute, chain multiple Text Decorator nodes with different placeholder tokens, each replacing one variable

How to Use

<aside>

1. Write your template → Add a Text Input node (or any text source) and write the template text with the placeholder where the dynamic content should go:

   Rewrite the following message as a professional email.

• Add a subject line
• Use a polite greeting and sign-off
• Keep all factual details exactly the same
• Tone: friendly but professional

Original message: "<<<input>>>"

1. Connect the dynamic source → Hook up whatever produces the variable text (OCR, LLM, file loader, etc.)

   

2. Wire both into Text Decorator → Connect the template to the top input port; connect the dynamic text to the bottom input port

   

3. (Optional) Customize the placeholder → If your template uses a different token (e.g., {{INPUT}}), enter it in the Placeholder token field on the node

4. Run Flow → Text Decorator outputs the template with the placeholder substituted; pipe it into an LLM, viewer, file output, or any downstream text consumer

   

   

   

   

</aside>