Plenio 1.5: Your Own Lyrics, Seen Templates and Creativity to 200
Plenio 1.5 for ComfyUI sings your own lyrics as written, writes style template values into the Song Brief, and lifts creativity to 200.
Plenio keeps the shape it had in 1.0: a brief, a local language model that writes the title, style and lyrics, a Song Sheet that shows exactly what the music model receives, YuE2-3B or MiniMax Music 3 performing it, then mastering and a release with tagged files, timed lyrics and sheet music. What 1.5 changes is how much of that path you can hold in your own hands.
What's new in 1.5
| Change | What it does |
|---|---|
| Your own lyrics | my own, sung as written in the Song Brief: the writer only fills in style, cover-art prompt and title |
| Templates you can see | 250 style templates now write their values into the brief's fields, which are lists of template values |
| Creativity to 200 | 100 becomes the default, 101-150 experimental, 151-200 adventurous |
| Freer covers | An original notes slider thins the source's instrument line and chords without writing anything new |
| Section tags of your own | The eight sections YuE2 learned as buttons, 8 more that Plenio fits to its plan, and tags you invent |
| Room for thinking models | Every LLM answer may be 16,384 tokens, with a 24,576-token context |
Lyrics you wrote, sung as written
The Song Brief with the lyrics choice on "my own, sung as written below", a verse and a chorus typed in.
A sung Song Brief now has a choice between written from the idea below, which is the default, and my own, sung as written below. In the second mode the writer writes only the title, the style and the cover-art prompt: the words stay exactly as typed, section tags included. Without tags, every block of lines becomes a verse and a block that comes back turns into a chorus, and the title is taken from the lyrics instead of from the genre. The same choice is in the Plenio Player's Make a song form, so a guest on a phone can paste lyrics too.
Templates that fill the brief
The genre field open: type your own at the top, then every genre the 250 templates know.
Choosing a template used to copy a block of text. In 1.5 it writes its values into the brief: description, genre, mood, tempo, key, meter, language, voice, theme, the lyrics idea and the lead instrument, plus its vocals, length and melody. A second template replaces only what a template put there and keeps what you changed, and an emptied field takes the template's value again. Every field is now a list of what all templates know, with ✎ type your own at the top for anything else. The templates' description says how a song sounds and a new lyrics field says what it is about.
Eleven templates were added for occasions and genres that had none: a birthday song in English and German, party Schlager, a German carnival song, a wedding first dance, a Liedermacher song, pop rock, melodic trap, Latin pop, J-rock and Italian pop, for 250 in all.
Creativity up to 200
Creativity's default is now 100, the experimental setting the author says gave the best songs, and every value up to 100 keeps the meaning it had before, so a saved 1.4 workflow at 50 still writes and samples at 50. Above that, 101-150 stays experimental with freer sampling, and 151-200 is called adventurous: a fresh blend with a clear touch of another genre, a signature instrument, or a surprising turn in the lyrics. At 200 YuE2's sampling is pushed to plan 1.0 / 0.96 / 64 and take 1.15 / 0.98 / 200, which the changelog calls the edge of what is reported to stay coherent.
Freer covers
The Cover Brief gains original notes (0-100, 100 by default): how many of the source's notes the score keeps. Lower values leave out notes of the instrument line and the chords step by step, down to roughly a quarter at 0, while the notes on strong beats and each section's first chord stay. Nothing new is written, so the cover cannot invent material it was not given. A sung cover's melody keeps every note so its words stay sung: the project measured 78 percent of the words heard at 0 against 83 percent at 100. The summary reports how many notes were kept.
Section tags you can define
The lyrics editor's buttons are now the eight sections YuE2 learned: [Intro], [Verse], [Pre-Chorus], [Chorus], [Bridge], [Interlude], [Instrumental] and [Outro]. A more button adds the tags Plenio fits to YuE2's plan ([Post-Chorus], [Hook], [Refrain], [Solo], [Break], [Breakdown], [Drop], [Build]), and + your tag keeps tags of your own such as [Violin Rise] or [Choir] for every song after. With words a custom tag is sung on its section; without words it becomes an instrumental part. A tag with characters a score's section names do not have no longer stops the run, which the changelog lists as a fix.
More room for thinking models
Every LLM answer may now run to 16,384 tokens with a 24,576-token context, up from 6,144, 2,048 and 1,024 in the writer's draft, the arrangement and the misheard-words check, in ComfyUI's native text generation and in the Local LLM node. That leaves room for a thinking model's reasoning before its answer, at the cost of a higher context requirement: an LM Studio model now needs at least 24,576. Saved workflows pick up the new lengths when they open, and Song API 1.3 exposes the new fields lyrics_from, lyrics, original_notes and creativity up to 200 while older API prompts run unchanged.
Getting Plenio 1.5
comfy node install comfyui-plenio-music. The pack needs ComfyUI 0.37.0 or newer and adds no Python packages of its own.
Check the machine
Open 0 · System Check from the template browser: it reports the ComfyUI version, the GPU and the model files each template still needs.
Pick a route
YuE2 · Song for a sung song from a brief, YuE2 · Cover to rework a recording, MiniMax · Song for a caption-driven render, or YuE2 · DAW to write the melody yourself.
The GitHub release for 1.5.0 is here, and the full list of changes is in the CHANGELOG.
Comments
Sign in with GitHub to join the discussion.