JSONL export

JSONL AI image dataset

Name: JSONL AI image dataset
Creator: Generated Gallery
License: Metadata index with provenance and source URLs. Check source rights before training or redistribution.

A free metadata-only AI image dataset packaged for scripts, agents, DuckDB, Python, and search experiments. Download compressed JSONL with prompt text, weak labels, source metadata, and schema links.

manifest.json full JSONL.GZ prompt-only JSONL.GZ GitHub

Compressed JSONL.GZ exports keep the dataset simple to mirror, diff, and process in batch jobs.

Every record keeps provenance fields so builders can review source context instead of blindly treating the web as soup.

Includes a prompt-only split for lower-risk analysis, prompt clustering, model evals, and search tooling.

Load it in one command

curl -L https://generatedgallery.com/index/generated-gallery.jsonl.gz -o generated-gallery.jsonl.gz
python examples/label_counts.py public/index/generated-gallery.jsonl.gz

The export is designed for lightweight analysis, search, prompt evaluation, weak-label audits, and downstream agent tasks. It is a metadata and prompt index, not an image ownership claim.

What is inside

Prompt text and negative prompts where available
Model family, source, category, tags, and safety flags
Weak labels for subjects, styles, composition, medium, and use cases
Provenance fields and source URLs for rights review
Prompt-only split for lower-risk analysis workflows

Built for builders who want data they can actually parse

Generated Gallery keeps the dataset public, versioned, and easy to inspect. If you build a loader, audit labels, or add a source adapter, open a GitHub issue or PR.