Sora 2 Cinematic Video Prompt (Three-Beat Micro-Film)
Sora 2 rewards a very specific shape: one subject, one action, one camera move, told in three timed beats. This prompt is built exactly that way — a lone hiker, a slow push-in, establish at 0:00, reveal at 0:03, payoff at 0:07 — with format, sound and grade written in, so the clip looks like a film still that moves. Two more micro-films on the same beats are inside.
The Sora 2 format that actually goes viral: one subject, one camera move, three timed beats — establish, reveal, payoff — written so the model holds the shot together.
Create a 10-second cinematic video in Sora 2, 2.39:1 anamorphic, 24fps, filmic grain, gentle vignette. One subject, one action, one camera move: a lone hiker in a dark coat stands on a misty mountain ridge at dawn, coat moving in the wind, layers of mountains fading into haze behind. [00:00] Beat 1 — establish: wide shot, the hiker small on the ridge, golden light breaking through fog. The camera pushes in slowly. [00:03] Beat 2 — reveal: medium shot; the hiker turns toward the light, the coat lifts in a gust, mist rolls through the valley below. [00:07] Beat 3 — payoff: the camera holds on the hiker's silhouette against the brightest sky, wind peaks, then the light softens. Natural sound only: wind, faint birds, soft rustle of fabric. No dialogue, no music, no text on screen. Warm teal-and-gold grade, subtle halation, realistic movement, no morphing, no extra people, no camera shake.
See what this prompt creates
Sora 2 made cinematic video easy to start and hard to finish well. Anyone can type "a beautiful mountain at dawn" and get ten pretty seconds. The clips that travel are the ones with a shape: something is established, something changes, something lands. That shape is not an accident of the model. It is written into the prompt, beat by beat, and this page is the template. It sits in our video collection alongside the Veo 3 product ad.
How to use it
- Paste the prompt into Sora 2 as written, with the time stamps intact. The stamps are the structure — removing them hands the pacing back to the model.
- Generate once and watch the beats, not the beauty: does the reveal land about a third of the way through, and does the payoff hold?
- To adapt it, change exactly one thing: the subject, the setting, or the camera move. Changing all three at once is how a working prompt stops working.
Why this prompt works
Four mechanisms do the work. The first is one of everything: one subject, one action, one camera move. Sora 2 drifts when a prompt asks for two actions or a cut. The second is the time-stamped beats — establish, reveal, payoff — each with one new piece of information, so the ten seconds have a beginning, a middle and an end. The third is the format block up front: aspect ratio, frame rate, grain and grade stated before the scene, which locks the look before the model starts inventing one. And the fourth is the sound and ban list: natural sound named, dialogue and on-screen text banned, which stops the model filling the silence with a narrator or a caption.
Two more micro-films on the same beats
- Coastal drone: keep the beats, swap the scene for "a lighthouse on a rocky headland at golden hour, waves breaking below" and the camera move for "a slow aerial arc around the lighthouse". Beat 2 is the arc reaching the sun side; beat 3 is the lift above the horizon into flare. Replace wind with surf in the sound line.
- Kitchen micro-drama: "a hand cracks an egg into a flour-dusted bowl in morning light", camera "slow handheld push-in, 35mm". Beat 2: the whisk blurs, steam rises behind. Beat 3: the hand stills on the yolk catching the light. Without beats, a close-up is just a screensaver.
Failure fixes: if the clip changes subject mid-way, cut back to one subject and one action and regenerate. If the beats blur together, give each beat exactly one new event. If movement looks rubbery, simplify the action — "coat moving in the wind" is one motion; "turns, walks, and waves" is three. If the grade comes back flat, restate the filmic grain and halation in the final line.
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
How long should a Sora 2 prompt be?
80–120 words is the working range for a final clip. Shorter prompts explore; the beat structure is for predictable results.
Do the time stamps actually work in Sora 2?
As pacing guidance, not frame-exact cuts. Expect the reveal near the stamp — judge the order of events, not the stopwatch.
Can I use this structure for a product video?
Yes — same skeleton as a commercial: establish, reveal detail, resolve on the hero shot. Keep one camera move and name the sound instead of music.
Why no dialogue or music in the prompt?
Dialogue is Sora 2's least reliable feature in short clips, and invented music often fights the edit. Natural sound keeps it clean; add a track yourself afterwards.
