Datasets:
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 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, 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
@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}
}
- Downloads last month
- 58