Research map / Image generation / Text-to-image diffusion
Text2QR: Harmonizing Aesthetic Customization and Scanning Robustness for Text-Guided QR Code Generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Text2QR generates aesthetic, text-customized QR codes by combining a Stable Diffusion/ControlNet pipeline with a blueprint image module (QAB) and latent-space refinement (SELR) that enforces scannability.
Problem
Aesthetic QR codes struggle to balance user-defined visual customization with reliable scannability; prior methods are style-transfer or halftone based and ControlNet-based approaches are unstable.
Approach
A QR Aesthetic Blueprint (histogram polarization, module reorganization, adaptive-halftone blending) produces a control image for ControlNet over Stable Diffusion, followed by iterative VAE-latent optimization with marker, code, and VGG-Wasserstein harmonizing losses to improve scan robustness.
Key results
- Scanning success rates stay at or above 96% across iPhone, Huawei, and TikTok/WeChat scanners at different sizes and angles.
- In a 24-subject user study over 200 QR images, Text2QR is preferred in most comparisons against ArtCoder and Quick QR (e.g., 89.2% vs 10.8% for one comparison).
- Text2QR achieves superior AesBench aesthetic scores over baselines.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Text2QR's QAB and SELR pipeline instead of prior aesthetic QR methods | QR code generation · Aesthetic QR codes | scanning robustness and visual quality | Helped |
| Remove Module Reorganization from QAB (ablation) | Text2QR QAB · Generated QR images | error level | Hurt |
| Remove Histogram Polarization from QAB (ablation) | Text2QR QAB · Generated QR images | error level | Hurt |
| Remove Adaptive-Halftone Blending from QAB (ablation) | Text2QR QAB · Generated QR images | error level | Hurt |
| Remove latent refinement (ablation) | Text2QR SELR · Generated QR codes | visual quality | Hurt |
| Remove marker loss (ablation) | Text2QR SELR · Generated QR codes | scannability | Hurt |
| Remove harmonizing loss (ablation) | Text2QR SELR · Generated QR codes | aesthetic quality | Hurt |
Limitations
- Scanning evaluation is limited to a small set of 20 codes on a single monitor setup and a few phones and apps.
- Refinement requires 400 optimization iterations per code, adding per-sample cost.
Experimental setup
Datasets: MS-COCO (VGG-19 pretraining)
Benchmarks: custom set of 100 generated 1024x1024 images AesBench
Metrics: scanning success rate (%) user preference (%) AesBench score
Baselines: QArt Halftone QR ArtCoder Quick QR
Builds on: Stable Diffusion ControlNet ArtCoder Halftone QR codes VGG-19
Closest papers by meaning
- Face2QR: A Unified Framework for Aesthetic, Face-Preserving, and Scannable QR Code Generation (NeurIPS 2024 · similarity 0.89)
- VQ-FONT: Few-Shot Font Generation with Structure-Aware Enhancement and Quantization (AAAI 2024 · similarity 0.83)
- Table2LaTeX-RL: High-Fidelity LaTeX Code Generation from Table Images via Reinforced Multimodal Language Models (NeurIPS 2025 · similarity 0.83)
- Image Regeneration: Evaluating Text-to-Image Model via Generating Identical Image with Multimodal Large Language Models (AAAI 2025 · similarity 0.83)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.83)
- Aesthetic Post-Training Diffusion Models from Generic Preferences with Step-by-step Preference Optimization (CVPR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.