Wikiprompt Is Now on Hugging Face: an Open Dataset of 56,000+ AI Prompts

Wikiprompt Is Now on Hugging Face: an Open Dataset of 56,000+ AI Prompts

Today we are publishing the Wikiprompt catalog as an open dataset on Hugging Face: **lautaschiaffino/wikiprompt-prompts**. This is our first dataset release (2026-08-18), a snapshot of the live catalog with 56,637 curated AI prompts for image, video and text models.

Why we are doing this

Prompting has become a real craft, but the best prompts are scattered across thousands of social posts and locked inside paywalled marketplaces. We think that knowledge should be open and shared, the way Wikipedia made reference knowledge open.

Wikiprompt is our attempt at that: a free, public, Wikipedia-style encyclopedia of the prompts people actually use to get great results from GPT Image, Midjourney, Seedance, Veo, Kling, Nano Banana, ChatGPT, Claude, Gemini, Grok, DALL-E, Flux and more. Every prompt is curated, categorized, tagged, attributed to its original author, and (for image and video prompts) shown next to the result it produced.

Publishing it as an open dataset is the logical next step. Following the Wikipedia model, the value is not in hoarding the data, it is in being the living, canonical, well-organized source. So we make the catalog openly accessible and let anyone build on top of it: research, apps, fine-tuning, analysis, or simply finding a great prompt.

What is in the dataset

One record per prompt, with the fields that make the catalog useful:

This first release is metadata only. The full prompt text lives on each prompt's page (the url field), and you can pull it programmatically too (see below). We will publish updated releases as the catalog grows, since it is updated daily.

Four ways to access the catalog

Code, docs and examples live in the open repo: github.com/lschiaffino/wikiprompt-dataset.

License and attribution

The compilation, curation, metadata, quality assessments and translations are licensed CC BY-SA 4.0, so you can reuse them with attribution and share-alike. The prompt bodies themselves are aggregated from public posts by their original authors, who keep the credit. When you use the data, please cite wikiprompt.org and, where shown, the original author.

Build something with it

If you are building a tool, running research, or training a model and a large, structured, open corpus of real-world AI prompts is useful, this is for you. Grab the dataset, open an issue on the repo, and tell us what you make.