Summary

Preparing worship lyrics for ProPresenter is often a repetitive and error-prone process. Song lyrics are commonly distributed as PDF or text documents, requiring volunteers to manually copy, clean, and format each song before it can be imported into ProPresenter.

This project automates the entire workflow using AI and n8n. Users simply upload a PDF or TXT file, and the system extracts, cleans, reformats, and returns lyrics that are ready to paste directly into ProPresenter.

The Problem

Many churches receive worship songs in PDF or plain text format. Before they can be presented during a service, someone must:

  • Open the document
  • Copy the lyrics
  • Remove unnecessary formatting
  • Split verses and choruses correctly
  • Reformat the text according to ProPresenter’s expected structure

While each song only takes a few minutes, the process quickly becomes repetitive when preparing multiple songs every week. Manual formatting also introduces inconsistencies and formatting mistakes.

Project Objectives

The project aimed to:

  • Eliminate repetitive manual formatting
  • Reduce preparation time for worship teams
  • Produce consistently formatted lyrics
  • Support both PDF and TXT files
  • Create a simple upload-and-download experience with minimal user interaction

The Solution

The workflow accepts either a PDF or TXT file through a simple upload interface.

After receiving the file, the workflow automatically extracts the text, sends it to an AI model for intelligent formatting, validates the output, and returns lyrics that follow a ProPresenter-compatible structure.

Instead of spending several minutes formatting each song manually, users receive a clean, ready-to-use version in seconds.

Technical Implementation

Workflow Components

  • n8n for workflow orchestration
  • AI (LLM) for intelligent lyric formatting
  • PDF text extraction
  • Text preprocessing and cleanup
  • File validation
  • Structured output generation

Workflow

Workflow 1

Workflow 1: Lyrics Formatting Pipeline

  1. Receive a PDF or TXT upload.
  2. Detect the type of upload (1 song per page (single PDF) or 1 song per file).
  3. Extract text from the document.
  4. Clean unnecessary whitespace and formatting artifacts and extract verses, choruses, and repeated sections using LLM.
  5. Send the extracted lyrics to an AI model with formatting instructions.
  6. Generate ProPresenter-compatible output.
  7. Return the formatted lyrics to the user.

Sample Output

Output

Challenges

One of the biggest challenges was handling the wide variety of input formats. Different PDF inputs have inconsistent text layouts, including irregular spacing, and formatting artifacts.

Prompt engineering was another challenge. The AI needed to consistently identify verses, choruses, bridges, and repeated sections while avoiding unnecessary modifications to the original lyrics.

Balancing flexibility with consistent output required multiple iterations of both preprocessing and prompt design.

Results

The workflow successfully automates a task that would otherwise require manual editing for every song.

Key outcomes include:

  • Reduced lyric preparation time from several minutes to a few seconds per song.
  • Consistent formatting across different input files.
  • Support for both PDF and TXT documents.

Lessons Learned

This project reinforced that AI performs best when paired with deterministic automation.

Rather than relying entirely on a single language model to extract lyrics, preprocess the document, and format the output, the system uses two models to clean and extract the text and generate the final formatted output. This approach allows for more flexibility and consistency in the output.

Future Improvements

Potential enhancements include:

  • Automatic extraction of song titles and metadata, with integration into the formatted output.
  • Support for songs that span multiple pages. The current system supports only two input formats: a single PDF where each page contains one song, or one song per file. Supporting multi-page songs would require more advanced preprocessing, prompt engineering, and improved song boundary detection.
  • Adding a storage layer to cache previously formatted lyrics, allowing duplicate requests to be served without reprocessing the original document.
  • Improve workflow logic to handle edge cases, such as empty PDFs or PDFs with no text. Also clean up the workflow to reduce complexity and improve readability.