Open AI image metadata
A public AI image dataset for prompts, provenance, and gallery packs.
Generated Gallery publishes a metadata-first AI image dataset: source links, prompt text where available, model names, tags, categories, weak downstream labels, safety labels, and JSONL/GZIP exports designed for search, research, agents, and small gallery experiments.
Metadata, not rights laundering
The dataset points back to source pages and upstream media. It is for discovery, prompt study, indexing, and provenance, not a claim that every image is free to reuse.
Weak labels included
Protocol v0.2 adds subjects, styles, medium, composition, use cases, model family, quality hints, and negative-prompt avoidance flags.
JSONL-friendly
Records are one object per line with compressed exports so consumers can stream, filter, validate, mirror, or feed the index into search pipelines.
Protocol v0.2
Labeled for downstream use.
The current labeler is deterministic and weak, but useful for filtering, sampling, prompt analysis, retrieval demos, and bootstrapping human or agent audits.
subjects
person, animal, product, landscape, architecture
styles
anime, photorealistic, cinematic, 3d-render, fantasy
medium
photo, illustration, painting, render, poster
composition
portrait, full-body, wide-shot, macro, isometric
use_cases
wallpaper, concept-art, character-reference, product-mockup
avoidance_flags
bad-hands, low-detail, watermark, anatomy issues
Dataset endpoints
Start with the manifest, then inspect samples before pulling the full feed.