ComfyUI LLM SDXL Adapter

A comprehensive set of ComfyUI nodes for using Large Language Models (LLM) as text encoders for SDXL image generation through a trained adapter.

<img width="1803" height="904" alt="image" src="https://github.com/user-attachments/assets/e8e5f047-37e7-4f8b-9bbd-78d70e2a7d80" />

Image with workflow

🎯 Available Adapters

RouWei-Gemma Adapter

Trained adapter for using Gemma-3-1b as text encoder for Rouwei v0.8 (vpred or epsilon or base).

Download Links:

• CivitAI Model
• HuggingFace Repository

📦 Installation

Requirements

• Python 3.8+
• ComfyUI
• Latest transformers library (tested on 4.53.1)

Install Dependencies

pip install transformers>=4.53.1 safetensors einops torch

Install Nodes

1. Clone the repository to ComfyUI/custom_nodes/:

cd ComfyUI/custom_nodes/
git clone https://github.com/NeuroSenko/ComfyUI_LLM_SDXL_Adapter.git

2. Restart ComfyUI

Setup RouWei-Gemma Adapter

1. Download the adapter:

   • Download from CivitAI or HuggingFace
   • Place the adapter file in ComfyUI/models/llm_adapters/

2. Download Gemma-3-1b-it model:

   • Download gemma-3-1b-it (non-gated mirror)
   • Place in ComfyUI/models/llm/gemma-3-1b-it/
   • Note: You need ALL files from the original model for proper functionality (not just .safetensors)

3. Download Rouwei checkpoint:

   • Get Rouwei v0.8 (vpred, epsilon, or base) if you don't have it
   • Place in your regular ComfyUI checkpoints folder

📁 File Structure Example

ComfyUI/models/
├── llm/gemma-3-1b-it/
│   ├── added_tokens.json
│   ├── config.json
│   ├── generation_config.json
│   ├── model.safetensors
│   ├── special_tokens_map.json
│   ├── tokenizer.json
│   ├── tokenizer.model
│   └── tokenizer_config.json
├── llm_adapters/
│   └── rouweiGemma_g31b27k.safetensors
└── checkpoints/
    └── rouwei_v0.8_vpred.safetensors

🔍 Debugging

To enable detailed logging, edit __init__.py:

# Change from:
logger.setLevel(logging.WARN)
# To:
logger.setLevel(logging.INFO)