---
tags:
- lora
- stable-diffusion
- sdxl
- controlnet
- qr-code
- qr-art
license: apache-2.0
base_model:
- SG161222/Realistic_Vision_V5.1_noVAE
- RunDiffusion/Juggernaut-XL-v9
---

# QR Art LoRA by QR-Verse

LoRA adapters trained on **gold-curated QR art images** for generating scannable, artistic QR codes using ControlNet QR Monster.

## Files

| File | Base Model | Size | Steps | Final Loss |
|------|-----------|------|-------|------------|
| qr_art_lora_sd15.safetensors | Realistic Vision V5.1 (SD1.5) | 13 MB | 3,735 | 0.22 |
| qr_art_lora_sdxl.safetensors | Juggernaut XL V9 (SDXL) | 89 MB | 1,005 | 0.50 |

## Training Config

- **Rank**: 16
- **Alpha**: 8
- **Dropout**: 0.05
- **Learning Rate**: 1e-4
- **LR Scheduler**: cosine_with_restarts
- **Target Modules**: to_q, to_k, to_v, to_out.0
- **Trigger Word**: 
- **Training Data**: Gold-curated QR art images (182 SD1.5, 67 SDXL)

## Usage

### ComfyUI (LoraLoader node)
- **lora_name**:  or 
- **strength_model**: 0.7
- **strength_clip**: 0.7

### Prompt Format
Prepend trigger word to your prompt:


## ControlNet Settings (CRITICAL)

- **Model**: [monster-labs/control_v1p_sd15_qrcode_monster](https://huggingface.co/monster-labs/control_v1p_sd15_qrcode_monster)
- **cn_start**: 0.0
- **cn_end**: 0.90
- **Sampler**: DPM++ 2M SDE Karras
- **Steps**: 50 (SD1.5) / 40 (SDXL)
- **CFG**: 7.5 (SD1.5) / 7.0 (SDXL)

## Part of [QR-Verse](https://qr-verse.com)