Sora 2 Prompts: 12 Copy-Ready Prompts by Shot Type
12 copy-ready Sora 2 prompts grouped by job: text-to-video, image-to-video, dialogue scenes and recurring characters, with the structure OpenAI recommends and the settings prose can't change.
By Eftal Yurtseven · Nov 18, 2026 · 8 min read

Most Sora 2 prompts ask for a movie and get a screensaver. The prompt says "cinematic, epic, beautiful city at night," the model fills every gap with its safest guess, and the clip looks like stock footage of nowhere. Good Sora 2 prompts read like a shot list: one subject, one camera move, a few timed beats, and the exact words anyone says out loud.
This post gives you 12 copy-ready prompts for Sora 2 on each::labs, grouped by the job you're doing, plus the structure behind them. The structure comes from OpenAI's own Sora 2 prompting guide; the settings come from the each::labs request schema, so you know which knobs live in the prompt and which ones don't.
The Settings Sora 2 Prompts Can't Change
Start here, because it saves the most wasted runs. Writing "make it 20 seconds" or "vertical video" in your prompt does nothing. OpenAI's guide is blunt about it: length and size are API parameters, and prose won't override them.
On each::labs that means four fields. duration takes 4, 8, 12, 16 or 20 seconds. aspect_ratio is 16:9 or 9:16. Sora 2 Pro adds a resolution switch between 720p and 1080p. And character_id lets you drop a saved character into the shot, which we'll get to below.
One more rule that belongs in this section: shorter clips follow instructions more reliably. OpenAI says that stitching two 4-second clips in the edit often beats one 8-second clip. So pick the shortest duration that holds your idea, and write a prompt sized for it. A 20-second prompt crammed into a 4-second clip is how you get a blur of half-finished actions.

The Four-Part Structure Behind Good Sora 2 Prompts
OpenAI's guide splits a strong prompt into four blocks, and every prompt below follows them.
Scene. Plain prose about who is there, what they wear, where they are, the weather, the time of day. Concrete beats pretty: "wet asphalt, zebra crosswalk, neon signs reflecting in puddles" tells the model far more than "a beautiful street at night."
Cinematography. The framing, the angle, the one camera move, and the mood. Style is the strongest lever you have here. "1970s film" or "16mm black-and-white film" decides a hundred small things at once.
Actions. A short list of beats in order. One camera move and one subject action per shot is the guide's rule of thumb, with the action described in timed steps ("takes four steps, pauses, looks back") rather than a vague verb.
Dialogue. Its own block, below the scene, with speakers labelled the same way every time. Keep it short. OpenAI's guidance is one or two short lines for a 4-second shot and a few more for 8 seconds.
Leave out anything that doesn't matter for the shot. An unspecified detail isn't a bug, it's room for the model to make a choice you didn't think of.

Text-to-Video Sora 2 Prompts
These four cover the jobs most people open the text-to-video tab for. Run them on 8 seconds first.
1. The documentary interview. A clear style plus one line of speech.
A 1990s documentary-style interview. An old fisherman in a faded yellow raincoat sits on an upturned crate inside a cluttered boat shed, nets hanging behind him, grey morning light through a dusty window. Medium close-up, eye level, locked-off camera, slight film grain and warm VHS color. He looks just off camera, smiles, and says: "The sea never cared who I was. That's why I liked it."
2. The product hero shot. One object, one move, controlled light.
Studio product shot of a brushed steel wristwatch resting on a slab of dark slate. A single soft key light from the upper left, a thin rim light tracing the bezel, black background. The camera does a slow 90 degree orbit around the watch at table height while a drop of water slides down the crystal. Macro lens look, crisp focus on the dial texture, calm premium mood. No text on screen.
3. The street at night. The guide's favorite example of concrete detail, turned into a full prompt.
Wet asphalt, a zebra crosswalk, neon signs in red and teal reflecting in puddles. A woman in a long charcoal coat crosses the street with a paper umbrella. Wide shot from a low angle, handheld with a slight sway. Beats: she steps off the curb, pauses halfway as a tram passes behind her, then keeps walking out of frame right. Palette: black, teal, red, warm sodium orange. Rain sound and a distant tram bell.
4. The hand-painted animation. Name the medium before anything else.
Style: hand-painted 2D animation with soft brush textures and warm tungsten light. A small round robot with one blinking eye tidies a cluttered inventor's workshop at night. Medium wide shot, static camera. Beats: the robot picks up a fallen screwdriver, places it in a jar, then notices a tiny potted plant and gently turns it toward the window. Quiet clockwork ticking, no dialogue.
Image-to-Video Prompts: Let the Picture Do the Describing
With Sora 2 Image to Video, your image becomes the first frame and the prompt only has to say what happens next. Two practical rules from the schema and OpenAI's guide: the image has to match the output size (1280x720 for 16:9, 720x1280 for 9:16 on the standard model), and you shouldn't re-describe what's already in the picture. Spend the words on motion. Our image to video prompts guide goes deeper on that habit.
5. The living product photo.
The camera slowly pushes in toward the perfume bottle. Morning light shifts across the glass and the liquid inside catches a warm glint. A few petals drift down behind the bottle and settle on the marble. Everything else stays still. Soft ambient room tone.
6. The portrait that breathes.
She blinks, takes a slow breath, and turns her head a few degrees toward the window as a strand of hair moves in a light breeze. Locked-off camera, no zoom. Keep her face, outfit and the background exactly as in the image.
7. The landscape with weather.
Clouds roll quickly over the mountain ridge from left to right, their shadows sliding across the valley. The lake surface ripples in the wind. Slow aerial drift forward, slight downward angle. Wind and distant birds.

Sora 2 Prompts With Dialogue That Lands
Sora 2 generates speech and sound with the picture, which is exactly why dialogue needs its own block. OpenAI's guide separates it so the model can tell what is said from what is seen. Label speakers, keep lines short, and size them to the duration.
8. The two-hander (8 seconds).
A 1970s romantic drama. Two people dance slowly on a rooftop at night, city lights blurred behind them, warm practical bulbs strung overhead. Medium two-shot, the camera circles them slowly. 16mm film look with soft halation.
Dialogue:
Ana: "See? Even the city dances with us tonight."
Leo: "Then don't let go."
9. The interrogation (8 seconds).
A detective drama. A cramped interrogation room, a single hanging bulb, cigarette smoke drifting. Over-the-shoulder shot from behind the suspect onto the detective, slow push in. Cold green and amber grade.
Dialogue:
Detective: "You're lying. I can hear it in your silence."
Suspect: (says nothing, looks down at the table)
10. The vertical ad line (4 seconds, 9:16).
A young barista behind a coffee counter looks straight into the camera, holds up a cup and grins. Bright, clean morning light, shallow depth of field, phone-camera look.
Dialogue:
Barista: "First one's always the best one."
Sora 2 Characters: One Face, Every Shot
Consistency across clips is the hardest part of any AI video series. Sora 2 handles it with characters. On each::labs, Sora 2 Characters takes a public MP4 of 2 to 4 seconds at 720p to 1080p plus a name, and returns a character ID. Pass that ID as character_id in any Sora 2 generation and, per the schema, write the character's name verbatim in the prompt. Characters can be animals or objects too, according to OpenAI.
11. The recurring mascot. Assume you saved a character named Pip.
Pip, a small orange tabby cat, sits on a kitchen counter next to a tall stack of pancakes. Medium shot, static camera, bright Sunday morning light. Beats: Pip stares at the pancakes, slowly stretches one paw toward the top one, then freezes when a door creaks off screen. Playful, warm mood.
12. The spokesperson, episode two. Same character, new set.
Maya stands in a small greenhouse surrounded by tomato plants, holding a watering can. Medium close-up, slow dolly left to right. Soft diffused daylight through the glass, palette of green, terracotta and cream.
Dialogue:
Maya: "Week two. The basil is winning."

Fix a Sora 2 Prompt in Small Steps
The fastest way to ruin a good clip is to change five things at once. OpenAI's advice is controlled, one-at-a-time changes that say what moves: "same shot, switch to 85mm." If you don't like the result, you know exactly which change caused it.
When a shot keeps failing, strip it down. Freeze the camera, cut the action to one beat, clear the background. Get that version working, then add complexity back one piece at a time. It feels slower. It is faster.
Two smaller habits pay off across a series. Name three to five palette colors in every prompt so shots cut together, and keep a fixed style line ("16mm film, warm tungsten") at the start of each one. If you're planning a lot of shots, our camera movement prompts and AI video prompt examples work as building blocks for Sora 2 too.
Standard or Pro for Your Sora 2 Prompts
Write the prompt once and decide the model by where the clip is going. Standard Sora 2 outputs 720p and is the place to test wording, timing and dialogue. Sora 2 Image to Video Pro and Text to Video Pro add the 1080p option, which is what you want for the final cut of a hero shot or an ad. A sensible workflow is to iterate on standard at 4 seconds, lock the prompt, then render the keeper on Pro at the length you need.
If you want to call Sora 2 from code, the same fields shown above go straight into a prediction request. Our Sora 2 API guide walks through the request.
Frequently Asked Questions
What makes good Sora 2 prompts?
Specific, visible details and a small number of moves. One subject action and one camera move per shot, the action broken into timed beats, a clear style line, and dialogue in its own block. Everything else can stay open for the model to decide.
How long should a Sora 2 prompt be?
As long as the clip, roughly. A 4-second shot needs a sentence or two of scene and one short line of speech at most. An 8 or 12-second clip can carry a few beats and a short exchange. Overloading a short duration is the most common reason actions get skipped.
Can I make a Sora 2 video longer by asking for it in the prompt?
No. Length is the duration setting (4, 8, 12, 16 or 20 seconds on each::labs), and the aspect ratio and resolution are settings too. Prose like "make it longer" is ignored.
How do I keep the same person across several Sora 2 clips?
Create a character from a 2 to 4-second clip with Sora 2 Characters, pass its ID as character_id, and use the character's name word for word in each prompt. Pair that with a fixed style line and palette and the clips will cut together.



