ComfyUI-SimpleTunerFlux2

ComfyUI nodes for using SimpleTuner-trained Flux 2 LoRA models.

SimpleTuner's Flux 2 architecture uses fused to_qkv_mlp_proj layers, making its LoRAs incompatible with standard ComfyUI Flux nodes. This package wraps SimpleTuner's pipeline directly.

Installation

cd ComfyUI/custom_nodes
git clone --recursive https://github.com/flowers6421/ComfyUI-SimpleTunerFlux2.git
cd ComfyUI-SimpleTunerFlux2
./install.sh  # Linux/macOS
# or
python install.py  # Windows

Nodes

Node	Description
Load Flux 2 Pipeline	Load base Flux 2 model
Load LoRA	Load SimpleTuner-trained LoRA
Sample	Generate images
Unload LoRA	Remove LoRA from pipeline
Select LoRA	Browse available LoRAs

Basic Workflow

[Load Flux 2 Pipeline] → [Load LoRA] → [Sample] → [Preview Image]

LoRA Paths

Place your trained LoRAs in:

• SimpleTuner/output/<project>/checkpoint-<step>/pytorch_lora_weights.safetensors
• ComfyUI/models/loras/
• Or use absolute path in the loader node

Requirements

• Python 3.10+
• PyTorch with CUDA
• ~24GB VRAM for Flux 2

License

MIT