Prompt pack

Gas Station Dance AI Prompts

Copy-ready prompts for recreating the Gas Station Dance trend in your own AI video tool: a second-by-second master prompt of the original clip, a prompt per part, four style twists and an avoid list. Free to use.

Updated · gasstationdance.org

The full timeline prompt

Use this with a model that accepts several reference images. Upload the dancer first, the selfie person second and the person in the car third.

Master prompt · 15 s · 9:16
Image 1 is the dancer, Image 2 is the selfie person who films, Image 3 is the person in the car. Keep each person's face, hair and outfit exactly as in their image; never swap or blend faces.
Raw vertical phone footage, handheld but smooth, early-morning light: blue sky with scattered clouds, low sunrise glow behind a red-trim gas station canopy, an empty asphalt lot, two dark luxury coupés, ground fog over green fields. Exactly ONE hard cut, at 4.3 s; every other transition is a fast whip pan or digital zoom.
0–0.5 s: wide, the dancer stands on the lot laughing and yelling at the camera. 0.5–2 s: drops into a deep wide squat, throws both arms overhead with clawed fingers, torso arched back, head thrown back — goofy, ecstatic, never freezing.
2–2.25 s: fast whip pan left. 2.25–4.3 s: front-camera selfie of the selfie person walking forward, grinning, other hand up in a hang-loose sign.
4.3 s: HARD CUT. 4.3–5 s: wide, the person in the car sits sideways in the open passenger door, a small white cup on the lap. 5–5.8 s: crash zoom to their face. 5.8–9.7 s: close-up, wind in the hair, serious look, then a smirk, a smile and an open laugh.
9.7–13 s: they stand up and groove gently between the two cars — hip sway, a little knee bounce, one step toward the camera, laughing.
13–13.4 s: whip pan up across the sky. 13.4–15 s: low-angle selfie, the selfie person with head tilted back, eyes closed, grinning.
Original instrumental: quiet before the cut, a heavy bass drop exactly at 4.3 s. Photorealistic, no text, no logos, no brand names, no license plates.

One prompt per part

For single-image models, generate each part separately and join them: dance + selfie (4.3 s), a hard cut on the beat drop, then the car part, then 2 s more of the selfie person. Each prompt refers to the one uploaded photo as Image 1.

Part 1 — the goofy dance (2 s)
Vertical phone video, wide shot of the person from Image 1 on an empty gas station lot at sunrise, blue sky with clouds behind. They laugh out loud, drop into a deep wide squat, throw both arms overhead with clawed fingers and arch back, head thrown back yelling at the sky — exaggerated, comedic, moving the whole time. Camera tilts up from a low angle. No text, no logos.
Part 2 — the laughing selfie (2 s)
Front-camera selfie video of the person from Image 1 walking across a gas station lot at sunrise, filming arm at the edge of the frame, grinning into the lens with the other hand raised in a hang-loose sign, red-trim canopy and low sun behind. Normal phone lens, no fisheye border. No text, no logos.
Part 3 — the car close-up (9 s)
Vertical phone video at a gas station lot at sunrise: the person from Image 1 sits sideways in the open passenger door of a grey luxury coupé holding a small white cup. Fast crash zoom to their face; wind blows their hair; a serious look turns into a smirk, a smile and an open laugh. Fast zoom out as they stand up and groove gently between two dark coupés, then step toward the camera laughing. No badges, no license plates, no text.

Style twists

Append one of these to the master prompt to change the mood without breaking the format.

Neon midnight
Swap the sunrise for midnight: deep blue sky, the canopy lit in pink and cyan neon, light rain, glossy reflections everywhere.
Desert highway
Set the station on an empty desert highway: dusty air, purple mountains on the horizon, the sun a red disc just above the hills.
Retro 80s VHS
Shot on a 1980s camcorder: soft VHS grain, slight chromatic aberration, no date stamp, warm faded colors.
Snowy forecourt
Light snow falling through the canopy lights, breath visible in the cold air, wet snow on the asphalt reflecting the low sun.

Avoid list

Paste at the end of any prompt
Avoid: extra people in the background, a fisheye or black circular border on the selfie, warped hands, face swaps between shots, brand logos or readable signs, license plates, captions or on-screen text, slow motion, camera shake that hides the face.

Tips that make the biggest difference

  • Bind every image to a shot. “Image 2 is the selfie person” beats “use these photos”.
  • Give timings. Seconds for every beat keep the hard cut near 4.3 s, where the trend sound drops, and stop one part eating the whole clip.
  • Name the light. Early morning — blue sky, low sunrise glow, ground fog — is what the original looks like.
  • Ban text and logos. Gas stations are full of signage; models love to invent brand names.
  • Use 9:16 and 15 seconds. That's the native size for TikTok and Reels.

Prefer not to prompt at all? Upload 1–3 photos here and we run the whole setup for you, or read the step-by-step guide.

FAQ

Which AI video models can use these prompts?

+

Any reference-to-video or image-to-video model that accepts photos of people and a text prompt. Models that take several reference images can do the whole clip at once; single-image models work best one part at a time with the single-part prompts.

How do I reference my photos in the prompt?

+

Most tools number uploads in order — “Image 1”, “Image 2” — while some use tags like @Image1. Upload the dancer first, the selfie second and the person in the car third, and keep that order in the prompt.

Why does my face change between shots?

+

Say explicitly which image belongs to which shot, ask the model to keep each face, hair and outfit exactly as in its image, and use one clear, front-facing photo per person.

Is there an easier way?

+

Yes — gasstationdance.org runs the whole thing for you from 1–3 photos, with no prompting, and cuts it on the beat drop.