Open dataset of tested AI image and video prompts

The Twistavo prompt dataset is a free, open list of 58 AI image and video prompts we tested on 12 models, each with its exact text, settings, credit cost and output.

Updated . Download CSV · Download JSON

What is in it

Prompts by model
ModelPrompts
Seedance 2.513
GPT Image 2.58
Nano Banana 2 Lite7
Seedream 5.0 Pro6
GPT Image 25
GPT Image 2.65
Kling 3.0 Standard3
Seedance 2.03
Z-Image Turbo3
Gemini Omni Flash 1.12
Seedream 4.52
Soul Cinematic1
Prompts by output
OutputPrompts
image37
video21

The first 20 of 58 rows

Newest pages first. The full prompt text is in the files and on each page.
ExampleModelSettingsCreditsOutputTested
Laundrette: a bodyguard presents one coin (Z-Image Turbo)Z-Image Turbo9:162image
Checkout: a weightlifter squat for one lemon (Z-Image Turbo)Z-Image Turbo9:162image
Wet sidewalk: a butler, an umbrella and a pigeon (Z-Image Turbo)Z-Image Turbo9:162image
Supermarket aisle: a recurring character portrait (GPT Image 2.5)GPT Image 2.52k, 9:1620image
Laundromat: a baguette held like a scepter (GPT Image 2.5)GPT Image 2.52k, 9:1620image
Hammerhead-shark bodybuilder, four views (Seedream 4.5)Seedream 4.53:29image
Pigeon athlete, four views (Seedream 4.5)Seedream 4.53:29image
Photographic person, face and full length (Soul Cinematic)Soul CinematicSoul Cinematic, text to image; Aspect ratio 4:3–image
Dog sidekick, full body and portrait (GPT Image 2)GPT Image 22k, 16:918image
Character bible, one image (Nano Banana 2 Lite)Nano Banana 2 Lite3:29image
Same prompt, clean labels (GPT Image 2.5)GPT Image 2.54k, 3:228image
Shot 1, the hook: a caravan on the dunes at duskSeedance 2.010 s, 720p, 9:16386video
Shot 2: the market from aboveSeedance 2.010 s, 720p, 9:16386video
Shot 3: the librarySeedance 2.010 s, 720p, 9:16386video
Still: the same library as an imageSeedream 5.0 Pro1k, 9:1611image
Still: the market at golden hourSeedream 5.0 Pro1k, 9:1611image
A man steps out of a café and takes a sipKling 3.0 Standard5 s, 16:954video
A woman cycles toward the camera in a parkKling 3.0 Standard5 s, 16:954video
A man walks down a sunny avenueKling 3.0 Standard5 s, 16:954video
An empty metro carriage on a viaduct, three seated-height viewsNano Banana 2 Lite1k, 16:99image

Download the files

Columns: page_url, example_id, example_title, prompt, model_name, model_id, duration_s, resolution, aspect_ratio, settings, studio_settings, credits, media_url, media_kind, date_tested.

How the rows are produced

  • One row per tested example of every published prompt page (the pages the prompts sitemap lists). A page with a single prompt and its video is one row.
  • A page whose examples use third-party input media (an uploaded source video, real photos of people or franchise material) is marked open_data: false in its data file and left out: its prompts stay on its page but are not redistributed here.
  • The prompt is the exact text sent to the model, copied from the page's data file, line breaks kept.
  • model_name is the catalogue name of the example's model_id. Older pages that predate per-example model ids keep the page's model name and leave model_id empty.
  • duration_s, resolution and aspect_ratio are the generation settings the example was quoted with; settings lists every setting as the page shows it; studio_settings lists the raw setting values.
  • credits is the cost of the example in Twistavo credits at those settings. A test recomputes it from the live price quote on every build, so a price change that moves it fails the build instead of leaving a stale number.
  • media_url is the output image or video as published on the page; date_tested is the date of the page that published the example.

License

License: CC BY 4.0. Reuse, share and adapt the data, also commercially, with this attribution: “Twistavo, https://twistavo.com/prompts”. CC BY 4.0 covers the prompt text and row metadata. Linked images and videos are not included in the license.

All prompt pages · Open AI Studio