SkyWhal3/ARIA_CVI_AAC_Photoset

---
license: cc-by-4.0
task_categories:
- image-classification
- image-to-text
language:
- en
tags:
- accessibility
- aac
- cvi
- cortical-visual-impairment
- assistive-technology
- communication
- special-needs
- pediatric
size_categories:
- 1K<n<10K
---

# CVI AAC Photoset

A free, photo-realistic image library for children with **Cortical Visual Impairment (CVI)** who use **Augmentative and Alternative Communication (AAC)** boards.

Cortical Visual Impairment is the leading cause of childhood visual impairment in the developed world. Existing AAC vocabulary libraries (ARASAAC, Mulberry, OpenSymbols, SymbolStix, Boardmaker, etc.) are pictogram-style — line drawings on white or colored backgrounds — which the child's cortical visual processing struggles to parse.

CVI users see best when shown:

- **Pure black backgrounds** (#000000) — eliminates visual clutter
- **Single, brightly-colored, high-saturation real-world objects** — saturated color is the strongest visual hook
- **Tight close-up framing** — fills the visual field of an AAC tile
- **Photo-realistic, not abstract** — supports object recognition and generalization to the physical world

This dataset is **1947+ images across 28 categories**, all rendered with one locked prompt prefix so visual style is consistent across the entire library.

## Contents

```
images/ ~1947 PNG files, 1024x1024
manifest.csv slug, label, category for every image
README.md this file
LICENSE CC-BY-4.0
```

## Methodology

- **Model**: black-forest-labs/FLUX.1-schnell (Apache 2.0)
- **Provider**: Together AI serverless inference (~$0.003/image)
- **Total compute cost**: ~$5.84 for the full dataset
- **Generation script**: included in the [ARIA Family Navigator](https://github.com/) repository at `scripts/generate-cvi-dataset.mjs`. Reproducible with one command + a Together AI API key.

### Locked prompt

Every image was rendered with this exact prompt prefix:

> A hyper-realistic, brightly colored, well-lit close-up photo of a single **{OBJECT}**. The object is large, perfectly centered, and fills most of the frame. Tight composition with minimal empty space around the subject. The background is pure, solid, pitch black #000000. No shadows, no gradients in the background. High contrast, clinical visual stimulus.

No post-processing (no `rembg`, no manual recoloring, no compositing) — the prompt itself guarantees consistent black backgrounds without halos or gradients.

## License

**CC-BY-4.0** — free for any use including commercial AAC apps, clinical materials, research, and remixing. Attribution requested.

FLUX.1-schnell outputs are unencumbered (Apache 2.0); this dataset license applies to the curation, prompt methodology, vocabulary list, and the dataset as a unified work.

## Acknowledgments

Built as part of [ARIA Family Navigator](https://aria-app.webflow.io/), a free desktop app for rare disease families. ARIA is a non-commercial parent-built project supporting STXBP1 epileptic encephalopathy, SNAP25 disorder, PEX10 / PBD spectrum, and the broader rare neurological community.

If this dataset helps your kid, your patient, or your project — that's why it exists. Use it freely.

## Citation

```bibtex
@dataset{cvi_aac_photoset_2026,
title = {CVI AAC Photoset: A Photo-Realistic Image Library for Cortical Visual Impairment AAC},
author = {Freygang, Adam},
year = {2026},
publisher = {HuggingFace},
license = {CC-BY-4.0}
}
```