Image, video, and text prompts — how we split the library

Prompy labels every prompt as image, video, or text. That label is not decoration. It tells you which kind of model the recipe was written for, and it is one of the first filters on Explore.

Image

Still frames: portraits, fashion, landscape, product, editorial. These recipes usually name a lens, lighting, film stock or sensor, and wardrobe. They are meant for tools such as Flux, Midjourney, GPT Image, and the image modes of ChatGPT or Gemini. The cover is a still that should match the direction.

Video

Motion recipes: camera moves, duration, weather, performance. They are written with Veo, Kling, Seedance, and similar tools in mind. A good video prompt says what happens over time, not only how a single frame looks. If you paste a video prompt into a still-image model you will get a poster, not a shot.

Text

Language recipes: system-style instructions, interview setups, script beats, caption frameworks. They belong in ChatGPT, Gemini, and other text models. The cover on a text prompt is a mood still so the card does not look empty. It is not the output.

Platforms on the card

Each prompt lists one or more platforms. That is a hint from the creator and the editor, not a lock. You can try a Flux-labelled portrait in Midjourney. You should not expect a Veo blocking-and-pacing recipe to sing in a still-image tool.

On Explore, filter by type first, then by look (Portrait, Fashion, Cinematic, and the rest). Sort by newest if you want the latest reviews, by trending if you want likes, or by most copied if you want recipes people actually took with them.

The split exists so the library stays readable. Three honest types beat a single pile labelled “AI.”