Plenio 1.5 in ComfyUI: Free Local AI Song Generator That Sings Your Own Lyrics

Plenio turns ComfyUI into a full local song studio. You write a short brief, a local language model drafts the title, style and lyrics, you approve a Song Sheet, and then YuE2 or MiniMax Music 3 renders the track. After that, Plenio masters and exports it with tags and cover art.

Plenio 1.5 came out on 1 November 2026. Its biggest new feature is your own lyrics sung as written. It also brings 250 style templates, a creativity dial up to 200, an “original notes” slider for covers, and much larger writer limits.

This guide covers the install, every model file with its exact folder and size, VRAM by card, the four routes, the new 1.5 features, licences (which matter if you want to sell songs), and fixes for common errors.

What Plenio Does

Most local music setups give you one node: a prompt goes in and audio comes out. Plenio wraps the whole production process around the model:

1. Brief: genre, mood, length (1:00 to 6:00), vocal language and theme.
2. Writer: a local LLM drafts the title, style tags and lyrics.
3. Song Sheet: you review and approve the plan before any audio renders.
4. Render: YuE2-3B or MiniMax Music 3 makes the track.
5. Refine: optional 48 kHz upsampling, stem separation and a vocal check.
6. Master and release: loudness, limiter, tags, cover art (FLUX.2 Klein 4B), and FLAC or MP3 export.

If you want to compare the underlying models first, my best local text-to-song models roundup covers YuE2, MiniMax Music 3 and the alternatives.

What Is New in Plenio 1.5

On covers, the project measured 78% of words heard with original notes at 0 and 83% at 100. A sung cover keeps every note of the melody.

Install Plenio

You need ComfyUI 0.37.0 or newer. Then pick one method:

• ComfyUI Manager: Manager, Custom Nodes Manager, search “Plenio Music Production System”, Install, then restart.
• Comfy CLI: comfy node install comfyui-plenio-music
• Git:

cd ComfyUI/custom_nodes
git clone https://github.com/jplenio/Plenio-Music-Production-System

Plenio installs no Python packages of its own. The only optional extra is faster-whisper, which covers with lyrics need:

python -m pip install faster-whisper

On the Windows portable build, use python_embeded\python.exe -m pip install faster-whisper from the portable folder.

After restarting, open 0 · System Check from the template browser and run it. It reports your ComfyUI version, your GPU and any missing model files.

Model Files, Folders and Sizes

Every file goes under ComfyUI/models/. Opening a template also triggers ComfyUI’s missing-model dialog, which can download them for you.

YuE2 route (templates 1 and 2)

MiniMax Music 3 route (template 3)

Optional extras (all routes)

For smaller cards, there are lighter swaps you choose in the loader nodes: minimax_music3_dit_int8_convrot (2.5 GB), flux-2-klein-4b-fp8 (4.1 GB) and qwen_3_4b_fp4_flux2 (3.8 GB). If you have 24 GB or more, there is also yue2_3b_bf16 (7.8 GB).

The Plenio docs do not say which folder these go in, so put each in the same folder as the file it replaces.

VRAM by Graphics Card

Only the YuE2 row was actually measured, on a 16 GB RTX 5060 Ti. The others are estimates from model sizes. The mastering step needs no GPU at all.

The Four Routes

Only cover songs you have the rights to. For royalty-free background tracks, my best AI music generator comparison covers the cloud options too.

Make Your First Song

1. Run 0 · System Check and fix anything it flags.
2. Open the YuE2 · Song template (template 1).
3. Fill in the Song Brief: genre, mood, a length such as 2:00, vocal language and a theme.
4. Pick a style template, or leave the style open.
5. Run. The writer drafts a Song Sheet, and the run stops at “waiting for approval”.
6. Open the sheet, edit lyrics or sections if you like, approve, and run again to render.
7. Listen, then render more takes if it ends early. Master and export as FLAC or MP3.

For a series of songs, switch the mode to “new song every run”. To try several takes of one song, use “one song, stop to review” and set the take seed to random.

Use Your Own Lyrics

This is the headline 1.5 feature. In the brief, set the lyrics source to your own text and paste your lyrics with section tags:

[verse]
City lights are fading slow
[chorus]
Hold on, hold on, we're almost home

Plenio keeps your words exactly and builds the style and arrangement around them. You can add your own section tags too, such as [bridge] or [outro]. Through the Song API, the same thing uses lyrics_from and lyrics.

Writer Model: Gemma 4, GGUF, LM Studio or Ollama

By default, the writer is Gemma 4 E4B running through ComfyUI’s native text generation. You can switch it to:

• A GGUF file in ComfyUI/models/LLM/, run with llama.cpp (set PLENIO_LLAMA_SERVER if it is not found).
• LM Studio: start its local server. In 1.5, the model needs a context of at least 24,576 tokens.
• Ollama: start the Ollama server, then press R in the writer node to refresh the list.

If you are not sure which local LLM tool to use, see my Ollama vs LM Studio vs llama.cpp comparison.

Licences: Can You Sell the Songs?

So songs made with the YuE2 routes are for personal, non-commercial use. If you plan to monetise music, read the MiniMax Music 3 licence carefully. Plenio writes the licences of the models used into each release record, which is handy for keeping track.

Troubleshooting

Plenio templates do not appear

Update ComfyUI to 0.37.0 or newer and check the startup console for import errors.

Red nodes or “missing node types”

Update ComfyUI. If System Check says a package such as av, PIL or scipy is missing, reinstall ComfyUI’s requirements in its own Python.

“Value not in list” for ckpt_name, unet_name or clip_name

The file is missing or has a different name. Place it in the folder from the tables above, or select your file in the loader.

The run stops at the Song Sheet

That is by design. Open the sheet, approve it and run again, or switch to “new song every run”.

Out of memory while rendering

Follow the System Check suggestions, switch to the int8 models, enable tiled decode for MiniMax, and keep songs shorter.

The song stops mid-phrase

YuE2 sometimes hits its length ceiling. Render another take.

A voice appears in an instrumental

Instrumental is a strong request, not a guarantee. Render several takes and use Check Vocals to keep the cleanest.

MiniMax: “more than 5 000 tokens”

Shorten the caption or lyrics by the amount the message shows.

Writer list only shows “(choose a model)”

Add a GGUF to ComfyUI/models/LLM, or start LM Studio or Ollama, then press R. If you see “answer was cut off”, raise max tokens and context, or turn thinking off.

“samples above full scale were clipped”

Keep the Plenio Master (or Loudness and Dynamics) node before Export.

My Take

Plenio is the most complete local song workflow I have seen in ComfyUI. The Song Sheet approval step sounds like extra work, but it saves you rendering full tracks from lyrics you would have thrown away anyway.

Version 1.5’s own-lyrics mode makes it useful for real songwriting, not just fun experiments. Just keep the YuE2 non-commercial licence in mind before you upload anything for money.

FAQ

Is Plenio free?

Yes. The code is Apache-2.0 and the models are free to download, but some model licences are non-commercial.

How much VRAM does Plenio need?

8-12 GB works with the int8 models and short songs, and 16 GB is comfortable. YuE2 was measured on a 16 GB RTX 5060 Ti.

Can Plenio sing my own lyrics?

Yes. That is new in 1.5: set the lyrics source to your own text and they are sung as written.

How long can songs be?

The Song Brief accepts lengths from 1:00 to 6:00. Shorter songs are safer on smaller cards.

Does Plenio work with Ollama or LM Studio?

Yes, for the writer step. Start the server, then refresh the writer node. LM Studio models need at least 24,576 context in 1.5.

Can I sell songs made with Plenio?

Not with the YuE2 routes, because YuE2 is CC BY-NC 4.0. MiniMax Music 3 has its own licence with commercial conditions, so read it first.

Which is better, YuE2 or MiniMax Music 3?

YuE2 is lighter (4 GB int8) and supports covers and the DAW route. MiniMax Music 3 is heavier but has its own sound and licence terms. Try both on the same brief.

Sources

• GitHub: jplenio/Plenio-Music-Production-System (README)
• GitHub: Plenio v1.5.0 release
• Plenio docs: Models and downloads
• Plenio docs: Troubleshooting
• ComfyUI Wiki: Plenio 1.5 news