AIDC-AI-ComfyUI-Copilot/LMSTUDIO_SETUP.md

LMStudio Integration with ComfyUI-Copilot

ComfyUI-Copilot now supports LMStudio as a local LLM provider. This allows you to run powerful language models locally on your machine without sending data to external APIs.

Setup Instructions

1. Install and Configure LMStudio

1. Download and install LMStudio from https://lmstudio.ai/
2. Download a compatible model (e.g., Llama 3.1, CodeLlama, or other OpenAI-compatible models)
3. Start the LMStudio server:
   • Open LMStudio
   • Go to the "Developer" tab
   • Click "Start Server"
   • Default server will run on http://localhost:1234

2. Configure ComfyUI-Copilot

1. In the ComfyUI-Copilot interface, go to the LLM configuration settings
2. Set the following parameters:
   • Base URL: http://localhost:1234/v1 (or your LMStudio server URL)
   • API Key: Leave empty (LMStudio typically doesn't require an API key)
   • Model: Select the model you loaded in LMStudio

3. Verify Connection

1. Click the "Verify Connection" button
2. You should see "LMStudio connection successful" if everything is configured correctly

Supported Features

• ✅ Model listing from LMStudio
• ✅ Chat completions
• ✅ Streaming responses
• ✅ All ComfyUI-Copilot features (workflow generation, debugging, etc.)

Common LMStudio URLs

• Default: http://localhost:1234/v1
• Custom port: http://localhost:YOUR_PORT/v1
• Network access: http://YOUR_IP:1234/v1

Troubleshooting

Connection Issues

• Ensure LMStudio server is running
• Check that the port (default 1234) is not blocked by firewall
• Verify the URL format includes /v1 at the end

Model Not Loading

• Make sure you have downloaded and loaded a model in LMStudio
• Check LMStudio console for any error messages
• Try restarting the LMStudio server

Performance Issues

• LMStudio performance depends on your hardware
• Consider using quantized models for better performance
• Adjust LMStudio's GPU/CPU allocation settings

Benefits of Using LMStudio

1. Privacy: All data stays on your local machine
2. No API costs: No per-token charges
3. Offline capability: Works without internet connection
4. Custom models: Use any compatible model you prefer
5. Full control: Adjust model parameters and behavior

Model Recommendations

For ComfyUI-Copilot, we recommend:

• Llama 3.1 8B Instruct (good balance of performance and quality)
• CodeLlama 7B/13B (excellent for code-related tasks)
• Mistral 7B Instruct (fast and efficient)

Choose models based on your hardware capabilities and requirements.