YANC_LMStudio

A custom node for a LMStudio integration into ComfyUI.

Updates

The new version is now using the LMStudio python SDK. With this it is possible to send images to vision enabled models. Unloading of models is now also possible for remote sessions.

Features

• Prompts can be sent to LMStudio via HTTP API endpoint.
• Handles 400 errors during API requests, attempting alternative approaches if needed.
• Supports system messages for structured prompting.
• Allows configuration of model identifiers and parameters like temperature and max_tokens.
• Includes optional model unloading in LMStudio and ComfyUI after usage.

Installation

To install the required dependencies browse to your custom_nodes folder inside of ComfyUI, open a command line and clone this repository or install via ComfyUI Manager:

git clone https://github.com/ALatentPlace/YANC_LMStudio.git

Please install the requirements.txt after cloning.

Usage

The node will put itself in the folders of my other node collection (YANC). You will then find it in the subfolder "LMStudio".

Configuration Options

• prompt: The input prompt to be sent to the AI model inside of LMStudio.
• image: Image input for vision enabled models.
• model_identifier: Identifier for the specific model used. Can be found in LMStudio.
• draft_model: Model for speculative decoding.
• system_message: A multi-line system prompt that sets instructions to the LLM.
• reasoning_tag: Tag used to identify reasoning sections.
• seed: Random seed (0 by default). Set it to fixed to prevent generations.
• ip: Hostname or IP address of the LMStudio service.
• port: Port number to connect to.
• temperature: Controls randomness in output, between 0.1 and 1.0.
• max_tokens: Maximum length of the generated response.

Model Handling

• unload_llm: Boolean flag to unload the LLM after use (default: False).
• unload_comfy_models: Boolean flag to unload COMFY models before sending a request to LMStudio (default: False).