ComfyUI-SimpleTunerFlux2
ComfyUI nodes for using SimpleTuner-trained Flux 2 LoRA models. SimpleTuner's Flux 2 architecture uses fused layers, making its LoRAs incompatible with standard ComfyUI Flux nodes. (Description by CC)
Nodes (5)
Your SimpleTuner Flux 2 LoRA won't load in the normal loader — use this one
Select LoRA (SimpleTuner Flux 2) — the dropdown that hands you a path instead of a loaded model
Unloading a Flux 2 LoRA without reloading the base model
Loading Flux 2 the way SimpleTuner needs it (and why the normal loader won't work)
Sampling a SimpleTuner Flux 2 pipeline — steps, CFG, and when to ignore the defaults
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.safetensorsComfyUI/models/loras/- Or use absolute path in the loader node
Requirements
- Python 3.10+
- PyTorch with CUDA
- ~24GB VRAM for Flux 2
License
MIT