AI Video Prompt Generator

An AI video prompt generator built around shots

A video prompt is really a shot list. This AI video prompt generator lets you describe a scene once — subject, place, style, light, colour and mood — then plan up to six shots, each with a shot size, angle, camera move and speed, lens, action, dialogue, sound effects and a length in seconds. A timeline shows how the shots add up.

The same storyboard is then written the way each vendor documents: one prompt per clip for Veo 3.1 with quoted dialogue, SFX and a separate negative prompt; [Shot 1: …] blocks for Kling 3.0 multi-shot with native audio; natural-language camera sentences and positive phrasing for Runway; camera “concepts” for the Luma API; and a clear generic layout for any other model.

The checklist catches what usually goes wrong: a Veo clip that is not 4, 6 or 8 seconds, a ratio Veo doesn’t make, dialogue too long for its shot, sound sent to a model that ignores it and “no …” wording that can backfire. For stills, use the AI image prompt generator.

How to write a video prompt shot by shot

1. Choose the model under Write the prompt for, or load an example.
2. Describe the Main subject once, then the location, visual style, pacing, light, palette and aspect ratio.
3. For each shot pick a shot size, angle and Camera movement, write what happens and add dialogue as Name: line.
4. Set the seconds per shot and use Add shot for more; watch the timeline.
5. Add ambient sound and anything to leave out, fix the checklist, then copy the prompt or the JSON.

AI Video Prompt Generator features

What each video model documents

Sources read on 14 October 2026: Google’s Veo 3.1 page in the Gemini API docs (updated 8 October 2026) and Google Cloud’s video prompt guide (updated 13 October 2026); Runway’s Text to Video and Gen-4 prompting guides; Kling’s prompt guide (7 August 2026); Luma’s Dream Machine API docs. Sora: OpenAI’s guide says the Sora 2 models and Videos API were shut down on 24 September 2026, so the generic layout is the closest replacement here.

Worked example: a two-shot storm scene

The start example plans an 8-second crane shot and a 6-second close-up with one line of dialogue. For Veo 3.1 it becomes two clip prompts and one negative prompt:

Shot 1 — 8 s, 16:9
Low-angle wide shot of a lighthouse keeper in a yellow raincoat on a rocky island during a night storm. The keeper climbs the iron stairs outside the tower as waves explode against the rocks. The camera cranes up slowly. Cinematic 35mm film look. The sweeping beam of the lamp and flashes of lightning, deep blue, slate grey and warm amber, tense mood. SFX: waves crashing, wind howling. Ambient noise: rain hammering on metal, a distant foghorn.

Shot 2 — 6 s, 16:9
Eye-level close-up of a lighthouse keeper in a yellow raincoat on a rocky island during a night storm. The keeper wipes rain from the lamp glass and stares out to sea. The camera stays locked off and still. Cinematic 35mm film look. The sweeping beam of the lamp and flashes of lightning, deep blue, slate grey and warm amber, tense mood. Keeper: "Hold on, whoever you are." Ambient noise: rain hammering on metal, a distant foghorn.

Negative prompt (all shots): text, watermark

Switch to Kling 3.0 and the same plan becomes one multi-shot prompt — the scene first, then [Shot 1: Low-angle wide shot, slow crane up] and [Shot 2: Eye-level close-up, static camera] blocks with the quoted line and an ambience line.

Directing tips that work across models

• One move, one action per shot — “takes four steps to the window and pulls the curtain” beats “walks around”.
• Use film words: dolly in, truck left, crane up, orbit, rack focus, shallow depth of field.
• Keep the subject’s description identical in every shot so the character stays recognisable.
• Short dialogue — about 2.5 spoken words a second, so an 8-second clip holds roughly 20 words.
• Say what you want to see — “a locked camera that remains still” rather than “no camera movement”.