Configuration
Plenio has very few settings. Everything else is chosen in the workflow.
Environment variables
| Variable | Values | Effect |
|---|---|---|
PLENIO_OFFLINE | 1 / 0 | never access the network; missing assets produce an error that names the folder and files to place |
HF_HUB_OFFLINE | 1 | Hugging Face's own offline switch; also puts Plenio offline (it cannot be undone with PLENIO_OFFLINE=0) |
HF_ENDPOINT | a URL | Hugging Face's own mirror setting: Plenio Setup fetches the catalogue's model files from the same paths on that host (each file is still checked against the catalogue's size and SHA-256; a token is never sent there) |
HF_TOKEN (or HUGGING_FACE_HUB_TOKEN) | a Hugging Face token | Hugging Face's own setting: Plenio Setup uses it for gated model files before a token saved with huggingface_hub; it is sent to huggingface.co only, never to a mirror, and never written into a file of Plenio's |
PLENIO_AUTO_DOWNLOAD | 1 (default) / 0 | fetch missing Plenio assets when a node that needs them runs; 0 gives an error with manual steps instead |
PLENIO_ASSET_DIR | a folder | where Plenio assets are stored (default: models/plenio) |
PLENIO_WORKER_TIMEOUT | seconds | how long an out-of-process worker may run in total (default 3600) |
PLENIO_WORKER_IDLE_TIMEOUT | seconds | how long a worker may run without reporting progress (default 300) |
PLENIO_LLAMA_SERVER | a program or its folder | the llama.cpp server that runs GGUF files for Local LLM (default: found on the PATH, in Unsloth Studio, winget or Homebrew); used only with the Studio add-on |
PLENIO_LLM_OTHER_APPS | 1 (default) / 0 | Local LLM also lists the model folders and running servers of other apps |
PLENIO_ADDONS | 1 (default) / 0 | 0 switches the Studio add-on off where it is installed (git clone, the release zip): Plenio then starts no other program - not ComfyUI from the player, no llama-server, no file manager -, like the Comfy Registry package (two ways to install) |
PLENIO_MODELS_PREFER | balanced (default), quality, speed, small | what auto chooses among a model's files that run on your card: the best one but keeping an installed file that is nearly as good; the best; the fastest; the smallest download (config file: [models] prefer) |
PLENIO_MODEL_PINS | component=file, comma-separated | these files instead of the plan's choice, e.g. yue2=yue2_3b_bf16.safetensors - a file name in the component's model folder, also a finetune of your own (config file: [models.pins]; the component ids are those of resources/models.toml) |
PLENIO_API_REMOTE | token (default), open, off | who on another computer the Song API (v1 and v2) answers: a program with a token made in Plenio Setup > API access; every computer, as before 2.0; nobody. This computer needs no token (config file: [api] remote) |
PLENIO_MODEL_DOWNLOADS | ask (default), auto, off | whether Plenio fetches a missing model file: when you confirm it in Plenio Setup; by itself when a run needs it; never - it then names the file and its folder (config file: [models] downloads) |
PLENIO_WORKER_DEVICE | auto (default), same, cpu, cuda:<n> | where Plenio's own workers (the lyrics ASR) run: the largest graphics card ComfyUI does not use, else ComfyUI's; ComfyUI's card; the processor; a card by number (config file: [workers] device) |
Invalid values stop with an error message instead of being ignored.
Config file
The same settings can be written to <ComfyUI user directory>/plenio/config.toml; environment variables win.
offline = false
auto_download = true
asset_dir = "D:/models/plenio"
[workers]
timeout_seconds = 3600
idle_timeout_seconds = 300
[llm] # Local LLM
llama_server = "D:/tools/llama.cpp" # the llama-server program or its folder (no other program is started)
other_apps = true # LM Studio, Hugging Face cache, Ollama ...
[llm.folders] # more GGUF folders: label = folder
"My GGUFs" = "E:/gguf"
[llm.servers] # more OpenAI-compatible servers: label = API base
"Workstation" = "http://192.168.1.20:1234/v1"Labels appear in the Local LLM list (My GGUFs · model.gguf, Workstation · model-id); they must not contain ·. Plenio stores no API keys; see Local LLMs.
Files Plenio keeps in the ComfyUI user directory
| Folder | What |
|---|---|
plenio/templates/ | your brief templates (Brief templates) |
plenio/arrangement/ | your creative modes, one Markdown file each (Creative modes) |
plenio/cache/llm/ | Local LLM answers: the same request with the same seed is answered from here without a call (delete the folder to forget them, or turn reuse answers off in the node) |
plenio/cache/asr/, plenio/cache/asr-notes/ | lyrics transcriptions and the editor's notes about them |
Unknown keys are reported as errors so that typos do not go unnoticed.
Model files
ComfyUI-format model weights (YuE2, SheetSage2, MiniMax, text models) are handled by ComfyUI: the templates declare the files, and ComfyUI offers to download missing ones. Plenio-owned assets (ASR models used by a worker, the default writer's Gemma 4 12B GGUF) are pinned to an exact revision and downloaded only when a node needs them, under the policy above.