Sora Prompts for Veo, Kling and Runway

Sora has closed. The prompts still work

OpenAI's help center says the Sora app and website were discontinued on April 26, 2026, and that the Sora API was scheduled to shut down on September 24, 2026. So as of October 2026 there's no Sora to paste these into. The format it made popular is still the right one for any text-to-video model, though: a short shot description with the camera, the motion, the length and the look written out. Every block on this page is written that way, and none of them depends on a Sora feature. Paste them into the video generator you use now. Our AI video generators page compares the ones still running, with prices.

If you made clips in Sora, the same help article explains how to export them and says the data is permanently deleted once any final export window closes. Read it before you count on old clips still being there.

What a video prompt needs

A photo prompt describes one moment. A video prompt describes a moment and what happens to it over a few seconds, and the model fills every gap you leave with something random. Six things close the gaps:

1. The opening picture. Subject, place and time of day, written as if it were a still photo. If the first frame is wrong, nothing that follows will rescue the clip.
2. One thing that moves. Waves break, steam rises, a fox turns its ears. Name the motion and its pace.
3. One camera move. Static, a slow push-in, a slider from left to right, a drone flying forward and tilting down. Give it a speed and say it's smooth.
4. Light and look. Where the light comes from, the palette, and a genre the model has seen plenty of: a nature documentary, a quiet drama, a camcorder home video.
5. Length and shape. The seconds, the aspect ratio (16:9 for a screen, 9:16 for a phone feed, 1:1 for a product tile) and the frame rate when it matters.
6. What to leave out. Text, logos, watermarks, recognizable faces. A closing «No…» line stops captions and signs you never asked for.

Put together, a pattern you can type from memory:

[Shot type] of [subject] in [place], [time of day].
[What moves, and how fast].
Camera: [one move], smooth and stabilized.
Light: [direction and quality]. Look: [palette or genre].
Duration [5-10] seconds, aspect ratio [16:9 or 9:16].
No text, no logos, no recognizable faces, no watermark.

If your generator has its own duration and ratio controls, set them to the same values the prompt names, so the words and the settings don't pull in different directions. If it makes audio, add one line such as «Sound: soft rain and distant traffic, no music». The blocks on this page are written silent, so leave that line out where there's no sound to control.

Which prompt for which clip

Each block below covers one kind of shot. Pick by what the clip is for, then fill the brackets.

A weak prompt, rewritten

This is the kind of request that usually gets typed into a video generator:

Beautiful drone shot of the ocean, cinematic, epic, 4K, highly detailed.

Nothing in it tells the model where the camera goes, what changes, or when the shot ends. «Cinematic» and «epic» describe a feeling without saying how to get there, and «4K» is an output setting that the prompt can't control. Here's the same idea as a shot description. We wrote it for this page; it follows the shape of the Coastal Drone Flyover:

Aerial drone shot over a rocky northern coastline in early morning.
The drone flies forward at a steady speed, low over the water, then
rises slowly to reveal a white lighthouse on the headland. Waves break
against dark rocks, thin mist hangs over the sea. Soft light from the
left, cool blues and grays. Smooth stabilized motion, no sudden turns.
Duration 8 seconds, aspect ratio 16:9. No text, no people, no boats,
no watermark.

Three changes do most of the work. The camera gets one route with an end point, the lighthouse, so the shot has somewhere to arrive. Something happens halfway through, the rise, which gives the eight seconds a reason to exist. And «cinematic» survives only as things a camera crew would actually decide: light from the left, a cool palette, stabilized motion.

Fixes for a clip that came out wrong

Most video tools don't hold a conversation the way a chatbot does, so the fix usually goes into the prompt itself before you run it again.

Before you rewrite anything, run the same prompt two or three times. Video models vary a lot between runs, and a prompt that failed once may be fine.

Where AI video still goes wrong

Every text-to-video model has the same handful of weak spots. Writing around them saves more renders than any clever wording.

• Hands and contact. Pouring, gripping, stirring and anything where two objects touch tend to melt or pass through each other. That's why most blocks here end with «no hands», and the Brand Mood Film Clip keeps hands in soft, slow motion.
• Continuity. The longer the clip, the more an object can drift: a boat changes size, a sign gains a letter. Shorter clips stay truer, and you can join several in any editor.
• Several actions in one clip. Ask for a door to open, a dog to run in and the camera to turn, and you'll get all three squeezed and half-finished. One action per clip, then cut them together.
• Words in the picture. Labels, signs and titles come out garbled. Add your own text in an editor afterwards, where you can spell it right.
• Real places by name. A famous landmark will be approximated. Describe the kind of place, as the City Dusk Time Lapse does with its bracketed skyline, unless an approximation is fine for the job.

Moving these prompts to another generator

The blocks are plain text, so they paste anywhere. Three adjustments make them land. If your tool's maximum length is shorter than the block asks for, cut the action down to what that length can hold. Squeezing ten seconds of motion into five tends to rush it. If the tool starts from an image, make the opening frame first with one of our AI photo prompts, upload it, and paste only the motion, camera and length lines. And if a block names a frame rate your tool doesn't offer, delete that phrase. We review Google Flow, Runway, Kling, Luma and Pika separately, and the AI video prompts page has more shot types written in the same shape.