Back · All projects
/04 · April 2026Open-source · launching

Morning Star Praise

Lyrics, sheet music, web resources, turned into worship slides and time-synced videos in one click.

/04 · MORNING STAR
/04 · MORNING STAR

// Premise

Free worship volunteers from the grind of last-minute media: from a paste of lyrics, a sheet-music image, or a YouTube URL, to ready-for-Sunday slides and a time-synced worship video. Shipping MIT open-source so any small church can produce a Sunday's worth of media in 30 minutes flat.

// Deep dive

The pipeline accepts any input: text, a YouTube URL, .pptx, PDF, sheet-music image. Each runs through a Vision LLM dispatch (seven providers, runtime-switchable: Gemini / GPT-4o / Claude / Qwen-VL / GLM / local Ollama) with PaddleOCR as fallback, cleanly separating lyrics from chords and notation. oemer handles OMR (Optical Music Recognition), cropping detected sheet-music staves onto each slide with a draggable lyrics textbox below. No re-rendering. The pixels are always your original upload.

Worship video uses faster-whisper + wav2vec2 for word-level alignment: upload MP3 + lyrics, and the rendered MP4 transitions slides at the exact moment each line is sung. Repeated choruses auto-expand. Karaoke mode highlights per-character (CJK) or per-word (Latin) in lockstep with the vocal. An embedded @remotion/player lets you nudge slide timings and swap backgrounds, then re-render from a cached plan. Two operating modes: pure-local (Ollama + faster-whisper, zero cloud dependencies) or LLM-powered (any of seven providers). Core pipeline runs at zero cost, zero data egress.

"…when the morning stars sang together and all the angels shouted for joy."

Job 38:7

// 01 · Origin

The 1 AM Saturday slide deck

Every worship team has a person called "the one making the PPT at 1 AM Saturday." Often that person was me. The same pain repeats every week:

  • "The lyrics are in Word — I just want a .pptx with our backgrounds and fonts."
  • "The worship leader sent a YouTube link; now I have to transcribe lyrics by hand."
  • "We have a printed sheet-music scan; we want each slide to show music + lyrics."
  • "We have an MP3 and want a karaoke / worship projection video."
  • "The song is bilingual (Chinese + English) and translating + formatting is brutal."
  • "We can't / won't send our church's data to a cloud LLM."

Five songs, four different sources, two hours of hand-assembly per Sunday. Commercial PPT tools don't understand Chinese-church needs (bilingual layout, custom backgrounds, liturgical flow). Morning Star Praise exists to turn those six "I want"s into one click — and to open-source it so every small church can benefit. The name comes from Job 38:7: "when the morning stars sang together and all the angels shouted for joy." The stars are the tools; worship is the purpose.

// 02 · Who it's for

Who this is for

  • Church worship teams — worship leaders, media volunteers, slide deck makers; the people pulling the late-Saturday shift.
  • Small / mid-sized Chinese churches — limited resources but plenty of creativity; need a bulk reduction in media-production friction.
  • Privacy-sensitive churches and ministries — those that can't or won't hand data to cloud LLMs.
  • Developer / volunteer hackers — want automation without vendor lock-in; able to fork and customize.
  • Cross-language gatherings — bilingual services, dual-language congregations, missionary outreach, all needing fast bilingual slide layout.

// 03 · What it solves

Compressing six input flavors into one pipeline

  • One unified entry — text, YouTube URL, .pptx, PDF, sheet-music image all flow through the same pipeline.
  • One unified output — a polished .pptx auto-styled with your background pool and fonts; the same React composition renders the browser preview and the final MP4.
  • Word-level alignment — MP3 + lyrics → video, with slides transitioning at the exact moment each line is sung.
  • Auto-expanded choruses — when a score writes the chorus once but the recording sings it three times, stanza matching replicates the lyric block automatically; no manual copy-paste.
  • Bilingual auto-layout — Chinese + English side by side on the same slide, translation routed to your chosen LLM.
  • Dual mode — pure-local (privacy-first) or seven-provider cloud LLM, switchable at runtime without restarting.

// 04 · Key Features

Word-level sync · 7 LLMs · 84 backgrounds · MIT open-source

  • 5 inputs: text paste · YouTube URL · .pptx · PDF · sheet-music image
  • 2 main outputs: a unified-style .pptx + a word-level-aligned MP4 worship video
  • Three-tier OCR fallback: Vision LLM (qwen3-vl / Gemini / GPT-4o) → PaddleOCR → oemer for sheet-music OMR
  • Word-level alignment: faster-whisper (large-v3) + wav2vec2 forced alignment
  • Karaoke mode: per-character (CJK) / per-word (Latin) highlight in lockstep with the vocal
  • Edit Video: embedded @remotion/player for nudging slide timings + backgrounds, then re-rendering from a cached plan
  • Background library: 84 bundled backgrounds (gradients, radial glows, motion landscapes) + user uploads
  • Seven LLM providers: OpenAI / Anthropic / Gemini / MiniMax / Qwen / GLM / Ollama, freely mixable, switchable at runtime
  • Pure-local mode: the core pipeline runs with no API keys (faster-whisper + wav2vec2 + PaddleOCR + Ollama)
  • Bilingual UI: Chinese default with English toggle

// 05 · Current Stage

Already in production internally · open-sourcing soon

  • Apr 2026 · Already in use across several churches as a production tool.
  • Core features stable: 5-input → PPT + video pipeline, word-level alignment, karaoke, dual mode, seven-provider LLM.
  • GitHub repo · github.com/leveo/morning-star-praise · MIT License in preparation.
  • Currently preparing: demo video, architecture blog post, one-click Docker deployment docs.
  • Targeted public open-source: 1.0 release with complete documentation and examples.

// 06 · Roadmap

Sheet-music Phase 2 · mobile · Garden integration

  • Sheet-music Phase 2: handwritten-score recognition (Phase 1 only handles printed scores) — the hardest piece.
  • Real-time rehearsal mode: upgrade Edit Video into a full live-worship remote (slide control, video volume, etc.).
  • Mobile: let media volunteers do 80% of the work from a phone (with the final 20% / preview on desktop).
  • Integration with Spiritual Growth Garden: "post-worship devotion" — auto-push this week's worship songs' verses into the Garden's daily reading.
  • Open ecosystem: invite contributors from other Chinese churches to ship backgrounds, templates, and bilingual translations.
  • Reverse-apply: repurpose the generated videos for church podcasts and social-media shorts.

// See it live

Reading is fine; trying it lives closer to the original intent.

View on GitHub