README.md

---
license: other
task_categories:
- image-to-image
- text-to-image
language:
- en
tags:
- lora
- flux
- image-editing
- style-transfer
- aura-style
- paired-image-dataset
pretty_name: Aura Style Paired Image Dataset
size_categories:
- n<1K
---

# Aura Style Paired Image Dataset

This dataset contains paired image examples for training a LoRA or image-editing model to apply a radiant aura lighting style. Each example has a conditioning/source image, a styled target image, and a prompt describing the target style.

## Dataset Structure

```text
.
├── conditioning/
│   ├── pair-001.png
│   └── ...
└── train/
    ├── metadata.jsonl
    ├── pair-001.png
    └── ...
```

`train/metadata.jsonl` contains one row per target image:

```json
{
  "file_name": "pair-001.png",
  "prompt": "A strong colorful light is illuminating man's silhouette...",
  "kind": "paired",
  "conditioning_path": "../conditioning/pair-001.png"
}
```

## Contents

- 25 paired training examples
- 25 target images in `train/`
- 25 conditioning images in `conditioning/`
- Image sizes:
  - 15 images at 1024 x 1024
  - 9 images at 1344 x 896
  - 1 image at 1486 x 1058

## Columns

- `image`: target/styled image loaded from `train/file_name`
- `prompt`: text prompt for the styled target image
- `kind`: record type; currently all rows are `paired`
- `conditioning_path`: relative path to the source/conditioning image
- `source_label`: optional provenance label for generated paired records

## Loading

```python
from datasets import load_dataset

dataset = load_dataset("ezhoureal/aura_style")
train = dataset["train"]
```

To resolve conditioning images:

```python
from pathlib import Path
from PIL import Image

row = train[0]
conditioning_path = Path("conditioning") / Path(row["conditioning_path"]).name
conditioning_image = Image.open(conditioning_path)
```

When using a local clone or `hf download`, the paths in `conditioning_path` are relative to `train/metadata.jsonl`.

## Intended Use

This dataset is intended for LoRA training and image-editing experiments where the model learns to transform a source image into a smooth, luminous aura style with colorful gradients, ethereal haze, contour lighting, and a cinematic glow.

The local training workspace for this dataset uses the folder root as an ImageFolder dataset:

```bash
accelerate launch -m lora.trainers.flux_lowmem \
  --dataset_name /path/to/training_data \
  --caption_column prompt
```

## Limitations

The dataset is small and style-specific. It is suitable for targeted LoRA training, smoke tests, and controlled style-transfer experiments, but it is not intended as a broad general-purpose image dataset.

## License and Usage Notes

The dataset card does not grant rights beyond the rights held for the included images. Verify that your intended use is compatible with the source image rights, model license, and any platform or training-service terms before redistributing derivatives or trained weights.