Today Plenio Music Production System reaches 1.0.0. I didn't pick the number lightly. Over the last three weeks Plenio went through fourteen releases, a lot of your feedback and every bug report I got. With 1.0 the whole way holds together – from a sentence describing a song to a lyric video on YouTube or a player you send to a friend – and the parts underneath have settled: workflows and songs you save with 1.0 keep working in every 1.x version.

And all of it is free: Apache 2.0 code, no subscription, no credits, no account, no cloud. It runs on your own GPU, inside ComfyUI's own nodes.

The idea: see exactly what the model gets

Most AI music tools are a text box and a slot machine. Plenio is the opposite. Before a single note is rendered, the Song Sheet shows you the title, style, lyrics and score the music model will receive. You can edit any of them, and your edits are never silently overwritten. If something changes underneath your edit, the run stops and asks.

What's in the box

Write. Song Brief and Cover Brief with templates for every genre. A local language model writes title, style and lyrics – Gemma 4 runs natively, or pick any GGUF from your models/LLM folder, LM Studio, Ollama, llama.cpp, vLLM or Jan. Only the writer you choose is loaded, and the GPU is free again before the music model starts. Thirteen creative modes (experimental) let the writer plan each section's chords, instrument lines and energy – Plenio writes and checks every note.

Compose. YuE2 sings from a score, and Plenio gives you a real score editor: notation, piano roll and ABC text, chord lane, bar inspector, playback with a sound per track, MIDI keyboard recording, arranging sections and bars like on a DAW's arranger track, the lyrics right over the notes they're sung on. MIDI in and out, MusicXML, sheet music as PDF/PNG/SVG. Or start from an empty score in YuE2 · DAW and write the melody yourself – YuE2 sings exactly what you approved.

Perform. YuE2 for songs, covers and your own melodies; MiniMax Music 3 for songs, with its 5,000-token budget checked before anything renders.

Cover. Any recording becomes a new version: SheetSage2 reads its score, faster-whisper its words. Keep the lyrics, write new ones fitted to the melody line by line, or go instrumental – and A/B against the original right in the editor.

Polish. Stems (vocals, drums, bass, other + a residual, so a neutral mix returns the song unchanged) with a mixer; Refine to 48 kHz; an EQ you edit on the curve; loudness to -14 LUFS / -1 dBTP with a true-peak limiter.

Release. FLAC 24-bit, MP3 V0 or WAV with tags and cover art (FLUX.2 Klein), the sheet music PDF, lyrics subtitles timed by listening, and a release record of everything that made the song – inside the FLAC. Drop that FLAC onto ComfyUI and you're back in its workflow: a new take, a changed score, new lyrics.

New in 1.0: the Plenio Player

Plenio used to stop at the exported file. Now there's the Plenio Player – a program of its own that opens in your browser, one click from ComfyUI (or a double-click without it). It finds every song you exported: search with filters like genre:jazz or bpm:>110, a page per song, the lyrics lit up syllable by syllable, the sheet music following the music, playlists, crossfades, A/B with the unmastered take. Open in ComfyUI takes any song back to its workflow.

And then the fun part:

🌌 Plenio Drop – I always loved Winamp's MilkDrop, so Plenio has its own renderer in that spirit, on your GPU: the feedback loop, warps, waves and shapes, 16 presets and one made up fresh every time, 60 fps in Full HD. Write your own as a short JSON file of numbers and small formulas ("zoom": "1.01 + 0.02 * bass") – save it and it changes on screen. It's data, never script, so a preset someone shares can only draw.

🎉 Live Play – a playlist for a party that you change while it plays. Make songs in ComfyUI during the evening and they join the moment Export Release writes them: next, at the end or anywhere. At the end it can simply wait for the next new song.

🎬 A song as a video – the song's visual, the lyrics as karaoke or subtitles, the notation scrolling along with the notes aglow, a title card. Full HD, 4K, square, or upright for Shorts, Reels and TikTok. The browser encodes it on your GPU – 12 seconds of Full HD in about 3 seconds here.

📦 Share your songs as a player – pick songs or a playlist, and Plenio packs the player and the songs (MP3, 128–192 kbit/s) into one HTML file. Send it: anyone double-clicks it in any current browser and gets the lyrics, the notation, the visuals and your playlists. No install, no server, no account – and no paths, workflows or FLACs from your computer.

🎤 Your look – 11 fonts and 11 effects for the lyrics (sweep, glow, bounce, neon, rainbow, on the beat …), in the player and in the videos. 📁 It also reads the folders you choose and plays your other music, with tags, covers and .lrc lyrics.

Also new in 1.0

• Lyrics subtitles timed by listening: (lines).srt and (notes).srt, measured by listening to the song and fitted to the score – half the words within 0.16 s of where they're sung. Ready for YouTube or DaVinci Resolve.
• Tags from the brief: Genre: {genre}, {artist} - {title}, {lyrics} as the lyrics tag – 43 variables.
• Plenio · Seed: one switch – 🎲 new seed or 🔒 same seed – and the number shown is always the one that made what you just heard.

Why 1.0 now

• The whole way is there. Writing, composing, covers, rendering, stems, mastering, export, continuing a song later – and now listening, video and sharing.
• It's stable underneath. 26 native nodes, the templates and the release record have settled; songs made with older versions open in today's templates.
• It's tested. 2,106 automated tests (Python, the frontend, and contract tests against a real ComfyUI 0.37.0), CI on Ubuntu and Windows, plus smoke runs with the real models.
• It's documented. A guide for every template and feature, and 9 narrated video tutorials recorded with the real models – every song you hear in them was made right there.
• It's honest. The experimental parts (creative modes, Stems, Refine) are marked as such; the limits are in the README.

Fine print

Plenio's code is Apache 2.0 and installs no Python packages. The model weights have their own licences and aren't included: YuE2 is CC BY-NC 4.0 (non-commercial), MiniMax Music 3 has its own community licence. Plenio's release record names the licences of the models each song was made with. Needs ComfyUI 0.37.0 or newer; YuE2 runs on 8–16 GB GPUs with the int8 checkpoint (the System Check template tells you what fits your card).

Get it

• ComfyUI Manager: search for Plenio Music Production System (comfyui-plenio-music)
• Comfy CLI: comfy node install comfyui-plenio-music
• By hand: git clone https://github.com/jplenio/Plenio-Music-Production-System into ComfyUI/custom_nodes
• GitHub (code, docs, changelog): https://github.com/jplenio/Plenio-Music-Production-System
• Tutorials: https://www.youtube.com/playlist?list=PLAFqTtP59fgE
• Promo (75 s): https://youtu.be/q_dMH834E1o
• Listen first: https://jplenio.github.io/Plenio-Music-Production-System/

Thank you to everyone who tried the early versions and reported what broke – 1.0 is built on that. I'd love to hear what you make with it. 🎶