New: GPT Image 2.5 Flare and Sunburst. Try them free with 3 images on sign-up, then 8 credits each.
Try GPT Image 2.5

Dataset methodology

Prompt Dataset Methodology

Methodology for the GPT Image Hub public prompt dataset, including fields, provenance, update cadence, and machine-readable distributions.

GPT Image HubUpdated 2026-10-21

Dataset scope

The public dataset contains prompt templates that are visible in the GPT Image Hub library and intended for discovery, retrieval, and reuse.

  • Each record includes full prompt text and canonical URL.
  • Each record includes category, tags, model defaults, aspect ratio, and attribution fields.
  • Translations are included when available so agents can map localized prompt intent.

Formats and discovery

The dataset is available in multiple machine-readable formats for search engines, AI agents, and data pipelines.

  • JSONL is the recommended format for bulk ingestion.
  • CSV is available for spreadsheets and BI tools.
  • A JSON Schema and manifest describe field semantics and distributions.

Provenance and freshness

Records expose source fields when known and include timestamps so agents can evaluate freshness and provenance.

  • The manifest includes version, updated_at, provenance, and same_as fields.
  • Manifest, schema, and dataset page responses include content-derived ETags; each record carries its own updated_at timestamp.
  • Category-level distributions allow smaller targeted crawls.

Public resources

GitHub and Hugging Face contain a licensed subset of the live library. Follow each record’s attribution and license; images have separate rights.