Artistic_Landscape

# Artistic Landscape Dataset

Welcome to the **Artistic Landscape**.

## Overview

**Artistic Landscape** is a curated collection of synthetically generated imagery designed to explore conceptual combinations across a wide variety of art styles, media, and visual characteristics. It is intended to be exploratory in scope, comparative across many aesthetic dimensions, and practical for downstream research workflows where a large, structured visual vocabulary is useful.

In total, we used **289** distinct media, **320** distinct styles, and **1,031** distinct characteristics to generate **50,000** samples.

## Generation Pipeline

### Image generation models

- `black-forest-labs/FLUX.1-schnell`
- `stabilityai/stable-diffusion-xl-base-1.0`

### Prompting models

- **Llama4-Maverick**
- **Gemini 2.5 Flash**
- **Gemini 1.5**

## Dataset Structure

Each example is a paired comparison record, with two images generated from the same prompt and criteria. The schema is:

- `pair_id` (string): Unique identifier for the pair.
- `image_a_model_id` (string): Model identifier used to generate `image_a`.
- `image_b_model_id` (string): Model identifier used to generate `image_b`.
- `image_a` (image): First generated image.
- `image_b` (image): Second generated image.
- `llm_prompt_generator` (string): Prompting model used to produce the prompt.
- `generation_prompt` (string): The full prompt text used for image generation.
- `timestamp` (string): Generation timestamp (as stored).
- `criteria` (struct): A structured set of categorical criteria used to define the conceptual combination for the pair, including:
- `abstract_styles` (string)
- `architectural_styles` (string)
- `artistic_movement_school` (string)
- `classical_and_ancient` (string)
- `color_palette_scheme` (string)
- `composition_layout` (string)
- `contemporary_movements` (string)
- `cultural_and_regional_styles` (string)
- `decorative_and_ornamental` (string)
- `decorative_arts_styles` (string)
- `digital_and_new_media_styles` (string)
- `emerging_and_experimental_styles` (string)
- `fashion_and_textile_styles` (string)
- `function_purpose` (string)
- `geographic_origin_cultural_context` (string)
- `illustration_and_commercial_styles` (string)
- `impressionism_and_post_impressionism` (string)
- `lighting` (string)
- `medium` (string)
- `modern_movements` (string)
- `mood_emotional_tone` (string)
- `perspective_viewpoint` (string)
- `photography_styles` (string)
- `printmaking_styles` (string)
- `psychological_and_emotional_styles` (string)
- `realism_and_naturalism` (string)
- `renaissance_and_early_modern` (string)
- `scale_size` (string)
- `sculpture_styles` (string)
- `subject_matter_content` (string)
- `technique_method` (string)
- `time_period_era` (string)

## Quickstart

### Download the whole dataset

This downloads the full dataset locally (large, ~131 GB).

```python
from datasets import load_dataset

ds = load_dataset("tabularisai/Artistic_Landscape")

example = ds["train"][0]
print(example.keys())

meta = {k: v for k, v in example.items() if k not in ("image_a", "image_b")}
print(meta)

img_a = example["image_a"]
img_b = example["image_b"]
```

### Stream a few samples (no full download)

This streams records and avoids downloading the full dataset.

```python
from datasets import load_dataset

ds = load_dataset("tabularisai/Artistic_Landscape", split="train", streaming=True)

samples = list(ds.take(5))
print(len(samples))
print(samples[0].keys())
```

## Limitations

- **Synthetic-only**: The images reflect the priors and biases of the underlying models and prompt distributions, and are not intended to represent historical art with scholarly accuracy.
- **Concept interactions**: Some combinations may be intentionally unusual or aesthetically conflicting, depending on how criteria interact.

## Citation
```bibtex
@misc{tabularisai_2026,
author = { tabularisai and Samuel Gyamfi and Vadim Borisov },
title = { Artistic_Landscape },
year = 2026,
url = { https://huggingface.co/datasets/tabularisai/Artistic_Landscape },
doi = { 10.57967/hf/8427 },
publisher = { Hugging Face }
}
```

## License

This dataset is released under the **Apache-2.0** license.

## Acknowledgements

This dataset was inspired by the *Image Preferences* work:
https://huggingface.co/blog/image-preferences