One prompt → one song → a two-scene music video (YuE2 × MiniMax H3)

# Score to Screen v3.6 Simple — Changelog (from v2.19)**One prompt → an original song + a 2-scene music video (YuE2 × MiniMax H3)**&gt; ### ⚠ v3 is a new, simpler line — not just the next v2 update&gt; v2.x grew into a "full control" workflow: 17 switches, several modes (H3-made music, song only, SFX only, frozen audio) and many fine settings.&gt; **v3.x Simple goes the other way.** It keeps only the everyday flow — prompt → song → 2-scene video — with **just 4 switches**. Everything else is decided automatically or fixed to the values that worked best in v2.**v3.x Simple (this release)**- Aim: easy, good results with little setup- Switches: 4 (Turbo, on-screen singing, drum sync, song delay)- Modes: song + video (YuE2 song, or your pre-made song)- Best for: most users, first-timers, batch runs**v2.34 Full / Legacy**- Aim: every option exposed- Switches: 17 + detail settings- Modes: also H3-made music, song only, SFX only, frozen audio- Best for: users who want to fine-tune or need the extra modes**Both are included and use the same custom_nodes**, so you can switch between them at any time. If you relied on a v2-only mode, keep using **v2.34 Full / Legacy**.---## 🆕 New in v3.3 – v3.6- 🎚 **A/V sync (v3.6).** H3's motion (hands, sticks) comes about **0.2 s after** the sound it hears. H3's own audio matches the song exactly; only the motion lags. - The final video's sound now plays 0.2 s later to match. - It is taken from 0.2 s earlier in the song, so nothing goes silent. Only a song used from its very start gets 0.2 s of silence first. - Adjust it in 🔧 Details → 🎚 A/V sync (0 = off, negative = sound earlier). H3's reference audio, the score and the cues are unchanged.- 📺 **Runs so far, inside the workflow (v3.4 / v3.5).** Each run adds one line, newest first: - the video file name, 🎬 video seed, steps, 🎤 vocal / 🎼 score seeds, and a 📝 memo - queue several runs and leave them; later you can see which video came from which seeds - the runs are also kept in <code>output/video/score2screen_run_log.csv</code> (opens in Excel), so the list continues after a restart- 🥁 **Clearer drum-sync messages (v3.3).** 📺 Drum hits now says why there is no hit table: - synced with Demucs (success), no drummer on screen, failed, or off- 🥁 **STRUM errors explained (v3.3).** - When STRUM makes no MIDI, the reason and a hint come first (e.g. demucs missing in STRUM's own Python). - The full log is saved to <code>output/score2screen_strum_cache/strum_last_log.txt</code>. - The setup guide now installs STRUM with <code>pip install -e ".[separation]"</code> and has a Troubleshooting section.## 🎵 v3.2: ▶ Start at for pre-made songs- Use your song from the chorus, or from anywhere you like. - <code>auto</code> (default): starts at the detected vocal onset, as before. - Seconds <code>62.5</code> or minutes:seconds <code>1:02.5</code>: starts exactly there.- The start is moved earlier only if both scenes would run past the end of the song (shown in 📺).- With a score, the bar-snapped scene cut, the playing cues and the lyric timing all follow the new start.## ✨ Highlights of the Simple redesign (v3.0 / v3.1)- 🎛 **Only 4 switches** (new Simple Board node): - ⚡ **Turbo**: fast generation (when off, 20 steps) - 🎤 **On-screen singing**: <code>auto</code> / <code>yes</code> / <code>no</code> - 🥁 **Drum sync**: <code>auto</code> / <code>off</code> - 🎵 **Song start delay** (seconds of silence at the start of the video)- 🎤 **"Song has vocals" and "the person on screen sings" are separate decisions** - A song can have vocals while nobody on screen lip-syncs (mouths stay closed). - <code>auto</code>: the LLM decides from your text and images. Scene 2 follows scene 1's decision. - The vocal / instrumental decision is made in this order: 1. strong instrumental words 2. vocals stated explicitly 3. "doesn't sing" alone means instrumental 4. otherwise the LLM decides- 🥁 **Drum sync <code>auto</code>**: the on-screen drummer follows the real drums of the song - Tries **STRUM** (neural drum transcription, optional), then **Demucs** drum-stem hits, then the **ABC score**. - Only runs when someone on screen plays the drums.- 🤖 **Now automatic / fixed**: - vocal onset detection (instrumentals start at 0 s) - cues for the other band members - first camera change no earlier than 3 s - extra style words for instrumentals - final audio is always the YuE2 song- 🎲 **Seeds** (video / vocals / score) are tucked into a collapsed "Seed details" group.- 🔧 **Rarely used settings** are grouped under "Details": - snapping the scene cut to bars - the song volume curve - A/V sync (v3.6)## 🗂 Removed from Simple (still in v2.34 Legacy)- H3-made music mode (fl2va) and its model / LoRA- Song-only mode, SFX-only mode, audio freeze- Experimental 8-class drum hit classification- YuE2 / H3 final-audio switch## 🎶 Notable improvements since v2.19**Pre-made songs**- Just pick a song and it's used automatically (YuE2 is skipped).- Choose where in the song to start with ▶ Start at (v3.2).- An <code>.abc</code> score with the same name is picked up automatically, even when the numbering differs.- 📺 shows which score was used.- Silence at the start of the song is measured, so no one "plays" during a silent intro.- Write "with vocals" or "instrumental" in scene 1's text to set the vocal type.**Band performance cues**- The splitter writes **who plays what** (e.g. "the blue-haired woman plays the guitar, the white-haired woman plays the drums").- Chord progressions are back in the rhythm cues.- Drums: hi-hat / snare / kick patterns, fills and crashes at section changes, descending tom fills.- Bass / keys move on chord changes; a second guitarist or drummer also gets cues.- Synced drum hits: kick and snare only, at most one per beat, up to 16 per shot.**Timing**- Vocal onset leaves room for both scenes.- The scene cut snaps to bar lines, and the onset check uses the same cut position.- First camera change is not before 3 s.- The final sound is lined up with H3's motion (A/V sync, v3.6).**Prompt quality**- Works around Civitai cutting off the first characters of the mp4 prompt.- No disclaimers in the H3 prompt.- Everyday words (car keys, oil drum…) are no longer mistaken for instruments.## 📦 Requirements / How to update- **custom_nodes v3.6** (included zip): - ComfyUI-PromptSplitter **v2.10** - ComfyUI-ABCTiming **v2.19** - ComfyUI-ABCScoreIO **v1.21** - ComfyUI-ABCInstrumentalize **v1.3**- Replace your old folders with these, then restart ComfyUI. Full install steps: <code>README.md</code> in the zip.- **STRUM is optional.** Without it, drum sync falls back to Demucs or the score. Setup guide: <code>custom_nodes/ComfyUI-ABCTiming/STRUM_setup_guide.md</code>- Each scene is up to 15 s. H3 runs twice, so it takes about 2× as long as the 1-scene version.- Both JA and EN versions of the workflow are included.

score2screen_2scene_v3.6_simple_en.zip

custom_nodes_en_v3.6.zip

v3.6

v7.0

v6.2

v6.1

v6.0

v5.9

v5.8.1

v5.7

v5.6

v5.4

v4.1

v3.9

v4.0

v3.7

v2.19

v2.09

Files with SHA256: dc4cf1498f5725f21da24fbaad38a912a33a3070a5fe02f47ea67c2d3f7d25a9