
# ♿ Colorblind-Safe Data Palette Swapper — Make Any Chart Accessible in One Prompt

A fine-tuned LoRA model built on **Qwen-Image-Edit-2511** that transforms the color palette of any data visualization into a colorblind-safe alternative — without changing a single data point, label, axis, or geometric element. Trained on a curated dataset of 34 image pairs spanning line charts, bar charts, heatmaps, stacked charts, and multi-series plots.

---

## ✨ What This Model Does

Over 300 million people worldwide live with color vision deficiency, yet the majority of data visualizations — dashboards, reports, research papers, infographics — are built with color palettes that are completely inaccessible to them. This LoRA solves that problem at the image level, with a single prompt.

Given any chart image, `make_cb_safe` remaps the color scheme to a perceptually safe palette — blue-yellow, blue-orange, grayscale with texture fills, or a universally accessible multi-color palette — while preserving every data point, label, gridline, and annotation exactly as they are.

Perfect for:

- ♿ **Accessibility compliance** — make dashboards and reports WCAG-friendly instantly
- 📊 **Research & academic publishing** — ensure your figures are readable by all reviewers
- 🏫 **Education & e-learning** — produce inclusive visual materials for students with CVD
- 🏢 **Enterprise reporting** — convert legacy charts to accessible formats at scale
- 🧩 **Neurodiversity support** — clear, high-contrast visuals for low-vision and neurodivergent users

---

## 🚀 Quick Start

### Trigger Word
Always include `make_cb_safe` in your prompt to activate the palette-swap behavior.

---

### 🔴🟢 Deuteranopia — Red-Green Deficiency (Most Common)
Affects ~6% of males. Red and green appear as similar shades of yellow-brown.

```
make_cb_safe Convert the color scheme of Image 1 to the safe blue-yellow palette in Image 2. All data points and labels must remain exactly as they are.
```

---

### 🔴 Protanopia — Red Deficiency
Affects ~1% of males. Red appears dark, and red-green distinctions are lost.

```
make_cb_safe Apply a protanopia color correction to Image 1 using the reference colors in Image 2. Do not alter any text, data, or geometric structures.
```

---

### 🔵🟠 Blue-Orange Safe — Deuteranopia + Protanopia Combined
Best general-purpose swap for the two most common CVD types.

```
make_cb_safe Replace the confusing red-green hues in Image 1 with the clear blue and orange palette from Image 2, preserving the chart's structure perfectly.
```

---

### 🔵🟡 Tritanopia — Blue-Yellow Deficiency
Affects ~0.01% of population. Blue and yellow are confused with green and red.

```
make_cb_safe Replace the chart colors in Image 1 with the color palette in Image 2, keeping all text, data, and geometric structures absolutely unchanged.
```

---

### ⬛ Grayscale + Texture — Print & Universal Accessibility
Best for printed materials, monochrome displays, or maximum inclusivity.

```
make_cb_safe Replace the chart colors in Image 1 with the grayscale palette in Image 2, and add distinct texture fills to each data area, keeping all text and geometry unchanged.
```

---

### 🌐 General — Unknown CVD Type
Use when you are unsure of the specific deficiency type.

```
make_cb_safe Map the colors of the chart in Image 1 to the accessible palette provided in Image 2 without modifying any geometric or text elements.
```

---

## 📥 Download

**With ModelScope Python SDK:**
```bash
pip install modelscope
```
```python
from modelscope import snapshot_download
model_dir = snapshot_download('isumenuka/colorblind-charts-v1')
```

**With Git:**
```bash
git clone https://www.modelscope.ai/isumenuka/colorblind-charts-v1.git
```

---

## ⚙️ Training Details

| Parameter | Value |
|---|---|
| Base Model | Qwen-Image-Edit-2511 |
| Training Type | LoRA (Image Edit) |
| Dataset Size | 34 image pairs |
| Trigger Word | `make_cb_safe` |
| Train Epochs | — |
| Repeat | — |
| Learning Rate | — |
| Optimizer | — |
| Network Rank Dim | — |
| Network Alpha | — |
| Resolution | — |
| Save Precision | — |

---

## 🎨 Sample Output

The model accepts a source chart and a reference palette image, then outputs the same chart with all colors remapped to the target palette. Supported transformations include:

- **Red-green → Blue-yellow** (deuteranopia safe)
- **Red-green → Blue-orange** (deuteranopia + protanopia safe)
- **Full color → Grayscale + texture fills** (print and universal accessibility)
- **General palette swap** (any source → any accessible target palette)

All geometric elements — bars, lines, points, grid, axes, labels — are preserved exactly. Only the color fill values are changed.

---

## 📝 Tips for Best Results

- Always start your prompt with `make_cb_safe`
- Provide a clean reference palette image as Image 2 — a simple color swatch or example accessible chart works best
- For heatmaps, add "without modifying any geometric or text elements" to prevent layout changes
- For bar and stacked charts, add "add distinct texture fills" for pattern differentiation alongside color
- Use the **Blue-Orange** prompt as your default if unsure — it covers the two most common CVD types
- Works across all chart types: line, bar, stacked bar, heatmap, scatter, multi-series

---

## 🌍 Why This Matters

This model was built for the **QwenImageLora Challenge on ModelScope — Track 2: AI for Good.**

Color vision deficiency affects approximately 1 in 12 men and 1 in 200 women globally. Despite this, most data visualization tools default to red-green color schemes that are invisible to a significant portion of the population. This LoRA is a step toward making data communication genuinely inclusive — without requiring designers or developers to rebuild their pipelines from scratch.

---

## 👤 Author

**isumenuka** — [ModelScope Profile](https://www.modelscope.ai/isumenuka)

---

## 📄 License

This model is released under the **Apache License 2.0**.


colorblind-charts-v1

Files with SHA256: 696e1c9d568a5552807ae466c0c4cbdafcbfe7b5f32536515f7920239be9be02