AI Dunk Video Prompt: Slam Dunk From One Photo
The AI dunk works because of the gap between the setup and the payoff: everyday clothes, a real arena, and then a flight no human could make. This prompt writes the whole highlight as timed beats — the takeoff, the slow-motion flight over the defender, the hang on the rim, the shout — with the face locked to your photo throughout, so the impossible dunk is unmistakably yours. Two-handed and windmill versions are inside.
The AI dunk: you, in your everyday clothes, taking off from the free-throw line at a packed night game, flying over a defender, and slamming it one-handed.
Create a 10-second vertical video, 9:16, in an image-to-video model, starting from the uploaded photo. Keep the person's face, hair, and build identical to the photo in every frame — same person throughout, no face changes. [00:00] A packed indoor basketball arena at night, bright court lights, crowd blurred in the stands. The person from the photo stands at the free-throw line in their everyday clothes — jeans, plain t-shirt, sneakers — holding a basketball, facing the hoop. A defender stands between them and the basket. Sound: crowd murmur, sneakers on wood. [00:02] The person takes off from the free-throw line, rising into the air in slow motion, knees up, ball held high in one hand, flying over the defender. Sound: the crowd gasp rising, one sharp sneaker squeak. [00:05] At the peak, the person slams the ball one-handed through the rim, hanging on the rim for a beat, looking down at the defender. Sound: the dunk thud, the net snap, the crowd erupting. [00:08] The person drops down, lands, and turns to the camera and shouts with a huge grin, chest out. Sound: crowd roar, one buzzer note. One smooth camera move: a slow push-in that follows the flight, no cuts. Realistic arena lighting, realistic ball physics, the ball clearly going through the net. No text on screen, no captions, no music beyond the crowd.
See what this prompt creates
The dunk clip is the purest one-photo video trend: nobody believes it, everybody wishes it were real. Its wave began in summer 2026 — a July template post passed 1.7 million views — and surged in October, when "ai dunk" became a Google autocomplete suggestion, at roughly 16 million views across TikTok, YouTube, and Instagram. What separates shared clips is the face holding to the rim. This prompt protects that, in our video collection beside the Sora 2 cinematic template.
How to use it
- Start with a clear, full-length photo of yourself in everyday clothes, standing, facing the camera.
- Paste the prompt into an image-to-video model with the photo attached, keeping the time stamps and face-lock line. Generate two or three takes: the first loses the face; later ones hold it.
- Judge the landing: the hang and the shout are the shareable second.
Why this prompt works
Four mechanisms do the work. The face lock first: identity is the constraint, the dunk the decoration. The everyday-clothes rule — jeans, plain t-shirt, sneakers — is the premise: a uniform makes a sports render; street clothes make it you. The timed slow motion makes the flight one beat, so the model stretches it. And the defender, flown over not run around, gives the joke its height.
Two more dunk clips
- Two-handed power dunk: keep the rest; replace the flight with: "takes off, rises in slow motion holding the ball in both hands above their head, slams it two-handed, hangs on the rim, knees tucked".
- Windmill dunk: replace the flight with: "takes off, swings the ball in a full windmill circle, and slams it through, the defender ducking".
Failure fixes: a changing face means a larger opening still plus "same face in every frame, no changes". Ball through the rim, not the net: simplify to "ball clearly passes down through the net, net snaps". Melting defender: "one defender in a dark jersey, clearly visible under the flight path". A uniform appearing: restore "everyday clothes, jeans and plain t-shirt, unchanged throughout" verbatim.
How to use this prompt
- Copy it — hit the copy button above to grab the raw prompt text.
- Fill the brackets — replace anything in [BRACKETS] with your own details.
- Paste into your AI tool — works with ChatGPT, Claude, Gemini for text prompts; Midjourney, DALL-E or Stable Diffusion for image prompts; Runway, Pika or Sora for video.
- Iterate — tweak one bracket at a time until the output is exactly right.
Frequently asked questions
Can I really dunk from one photo?
As a video, yes — the image-to-video model animates your photo through the flight. The face lock in this prompt is what keeps it recognisably you instead of a generic player in your clothes.
Why everyday clothes instead of a basketball kit?
The joke is the gap: a normal person, in normal clothes, doing something impossible. A kit makes it look like a sports render and the clip loses its whole premise.
What if my face changes halfway through the dunk?
That is the format's classic failure. Start from a bigger, clearer full-length photo, restate the face lock, and generate again — the second or third take usually holds it to the rim.
Does this work in Sora 2, Kling, or Veo?
All three handle it. Sora 2 holds the face best through the flight in our tests; Kling renders the crowd and arena light more richly. Generate in both and keep the stronger highlight.
