Orpheus TTS for ComfyUI (via LM Studio)

A custom node for ComfyUI that generates high-quality, natural-sounding speech using Orpheus TTS models running in LM Studio.

I tried all the other Orpheus nodes for ComfyUI, and they were either slow (2+min for 15s), or I couldn’t get them to install right. Unlike those, this doesn’t try to run the LLM model with all the different settings via some cli thing, I just rely on LM Studio where you can handle all the loading there. I got this to be as fast as I could, about 0.5x realtime on a RTX3090. Voices sound so good.

And now for your regularly scheduled LLM generated readme:

This node is designed for performance and user experience, featuring a highly-responsive, parallelized workflow. It streams audio tokens from LM Studio while decoding them into sound in a separate thread, ensuring your ComfyUI interface remains smooth. The state-of-the-art progress bar intelligently displays the real-time progress of the entire operation, not just one part of it.

*(Note: You will need to replace this with an actual screenshot of the node in your workflow.)*

✨ Features

High-Quality Speech Generation: Leverages the power of Orpheus TTS foundation models for expressive and natural voice output.

Seamless LM Studio Integration: Connects directly to your local LM Studio server, allowing you to use any compatible Orpheus model you have downloaded.

🚀 Blazing-Fast Parallel Processing: The node fetches audio tokens from the LLM and decodes them into audio simultaneously, maximizing performance and reducing wait times.

Full Control: All standard LLM inference parameters (temperature, top_p, seed, etc.) are exposed for fine-tuning your results.

✅ Requirements

ComfyUI: A working installation of ComfyUI.

LM Studio: The LM Studio application installed and running.

An Orpheus Model: You must have an Orpheus-compatible GGUF model downloaded in LM Studio.

We recommend starting with isaiahbjork/orpheus-3b-0.1-ft-Q4_K_M-GGUF. You can search for and download this directly from the LM Studio app.

Python Dependencies: This node requires the lmstudio and transformers Python libraries.

⚙️ Installation

Navigate to your ComfyUI custom_nodes directory:

“`bash

cd ComfyUI/custom_nodes/

“`

Clone this repository:

“`bash

git clone https://github.com/phazei/ComfyUI-OrpheusTTS-LMStudio.git

“`

*(Note: Replace the URL with your actual repository URL after publishing.)*

Install the required Python packages:

“`bash

cd ComfyUI-OrpheusTTS-LMStudio

pip install -r requirements.txt

“`

*(You should create a requirements.txt file in your repo with the following content:)*

“`

lmstudio>=0.2.20

transformers

torch

numpy

“`

Restart ComfyUI.

🚀 Setup & Usage

1. In LM Studio

Download Model: In LM Studio, search for and download an Orpheus model (e.g., orpheus-3b-0.1-ft).

Load Model: Select the model to load it.

Start Server: Navigate to the “Local Server” tab (the <-> icon) and click Start Server. This is the crucial step that allows ComfyUI to communicate with the model.

2. In ComfyUI

Add the Node: Right-click on your graph, select “Add Node,” and find the OrpheusTTS/LM Studio > Orpheus TTS (LM Studio) node.

Enter Your Text: Type the text you want to convert to speech in the text field.

Select a Voice: Choose one of the built-in English voices from the voice dropdown.

Set the Model Key (Optional but Recommended):

The model_key field tells the node which specific model to use in LM Studio.

To find the key, look at the top of your LM Studio server window. It will show the path-like identifier for the loaded model (e.g., orpheus-3b-0.1-ft).

Copy this identifier and paste it into the model_key field. If you leave it blank, the node will use whichever model is currently loaded in LM Studio by default.

Generate: Queue the prompt. The node will connect to LM Studio, generate the audio, and output an AUDIO artifact that can be used with video nodes or saved.

Troubleshooting

Node not appearing: Ensure you have restarted ComfyUI after installation.

Connection Errors:

Verify that the LM Studio server is running.

Check that your system’s firewall is not blocking the connection on the port LM Studio is using (usually 1234).

Slow Performance: Enable the debug flag on the node and check the logs in your console. The performance timers will show you whether the slowdown is happening during the LLM phase or the SNAC decoding phase.

License

This project is licensed under the Apache 2.0 License