Sora 2 prompts that still work in Veo, Kling and Seedance

Most Sora 2 prompt lists were written while Sora was live. This page covers what to keep, what to rewrite and what to drop when you move a Sora 2 prompt to Veo, Kling or Seedance.

Everything here comes from the model makers' own guides, read on 2026-10-26. See Sora alternatives for the dates and what replaced each Sora strength.

Do Sora 2 prompts still work now that Sora is retired?

Sora 2 prompts still work as text in other video models, but the Sora 2 model no longer runs. OpenAI's deprecation table lists the Videos API, sora-2, sora-2-pro and their dated snapshots as removed on 24 September 2026, with no replacement named.

A Sora prompt is a scene description. Scene, camera, action, lighting and sound are things every current model reads. The parts that break are the ones tied to Sora's own settings: clip length, cameos and character reference clips.

What made a good Sora 2 prompt?

OpenAI's guide treats a Sora 2 prompt like a storyboard sketch: camera framing, depth of field, action in beats, and lighting with a palette. It adds three rules.

• Keep one camera move and one subject action per shot.
• Count actions out, for example "takes four steps to the window, pauses".
• Put dialogue in its own block under the scene text, with short lines.

OpenAI also says shorter prompts leave the model more freedom, and longer prompts give more control that it may not follow. The same prompt gives a different clip each run, which the guide calls a feature.

How do Sora 2 prompts transfer to Veo, Kling and Seedance?

Scene, camera and action text transfers to all three models, and dialogue formatting and clip length are what you rewrite. The table compares each Sora element with the vendor guidance read on 2026-10-26.

The last row is where Sora cameos have no direct match. Google's guide documents reference images for Veo. Kling's guide describes video element references (3 to 8 seconds), multi-image elements with audio, and element voice control. fal's Seedance guide does not cover reference syntax, so read the current Seedance docs before you build around it.

Which popular Sora prompt types transfer, and which break?

Prompt types that rely only on a scene and a camera transfer, and types that rely on Sora features or named characters need a rewrite.

How to rewrite one Sora 2 prompt for three models

The example below is an original prompt, built from each vendor's syntax rules.

Sora 2 version (8 seconds):

90s camcorder look. An old Swedish fisherman sits on an upturned crate by a harbor. Cinematography: medium close-up, static camera. Actions: he looks at the lens, then at the water. Dialogue: Fisherman: "The sea gave me more than I asked for." Sound: gulls, rope creaking.

Veo 3.1 rewrite (cinematography first, speech in quotation marks, labeled sound):

Medium close-up, static camera, 90s camcorder look. An old Swedish fisherman sits on an upturned crate by a harbor. He looks at the lens, then at the water, and says, "The sea gave me more than I asked for." Ambient noise: gulls and creaking rope.

Kling 3.0 rewrite (speaker named, line next to the name, delivery noted):

An old Swedish fisherman sits on an upturned crate by a harbor, 90s camcorder look, medium close-up, static camera. He looks at the lens, then at the water. The fisherman says quietly, with a tired smile: "The sea gave me more than I asked for." Sound: gulls and a creaking rope.

Seedance 2.0 rewrite (subject and motion first, double quotes, sound named, music excluded):

An old Swedish fisherman sits on an upturned crate by a harbor. He looks at the lens, then at the water, and says, "The sea gave me more than I asked for." Medium close-up, static camera, 90s camcorder look. Gulls and creaking rope, no music.

What the length limits do to a 20-second Sora prompt

A 20-second Sora 2 clip becomes three Veo clips of 8 + 8 + 4 seconds, or two 10-second Kling or Seedance clips. Generate every clip at its maximum length and you trim the extra. A 16-second Sora clip needs two Veo clips at 8 seconds. OpenAI's own guide already suggests stitching short clips, because the model follows instructions better in shorter ones. Use the AI video cost calculator to price the extra clips.

Do I need a Sora 2 prompt generator?

A Sora 2 prompt generator is not required, because the Sora structure fits in four lines: scene, cinematography, actions and optional dialogue. A generator saves typing.

If you use one, pick a generator that supports your target model's clip lengths and dialogue format, and check its output against the table above.

Are viral Sora 2 prompt lists safe to copy?

Viral Sora 2 prompt lists are leads, not proof. One published Sora prompt list says its 20 prompts "have been tested" and drive high view counts, but gives no data or method for either claim. Its own dates are inconsistent, and it does not mention the shutdown.

Several prompts on such lists name real franchises or place a real person through a cameo. Rewrite both before you publish on any platform.

Likeness and copyright: what changes outside Sora

Sora 2's cameo feature put a user's own likeness into clips, and no other model in this comparison has the same feature in the guides read here. Do not put a real person's face into a prompt without their permission. Do not copy a named film or cartoon scene into a commercial video. Each platform sets its own rules; check the terms of the tool you use.

For ready-made prompts by format, see these AI video prompt examples. If Veo returns a silent clip, fix missing Veo 3 audio. For model choice, see the AI video model finder. For Kling-specific work, the Kling pricing guide lists current costs.

Which Veo endpoint to use

Google's Veo 3.1 guide is dated 16 October 2025. Google's Gemini API page now recommends Gemini Omni Flash and says the Veo 3.1 preview there shut down on 22 October 2026. Check which Veo or Omni endpoint your tool uses. The Seedance column follows fal's guide.