AI video prompting: a shot-by-shot prompt guide
Good AI video prompts describe one shot at a time, the way a director briefs a camera crew: subject and action, setting, framing and camera movement, lighting, style and mood, and sound if the tool supports it. Plan a shot list first, keep style words consistent across shots, use reference images for characters and products, generate a few takes per shot and change one thing at a time when refining. The same approach works in Veo, Kling, Runway and most other video models.
Why do AI video prompts fail?
Video models fill gaps with guesses. A prompt such as “a video about a restaurant” leaves the model to decide the subject, action, framing, lighting and style, so results look random and change every time. Describing a single shot precisely gives the model far less to guess. This guide applies to the main tools covered in our AI video generators comparison, including Veo, Kling and Runway.
What should every shot prompt include?
| Part | What to write | Example |
|---|---|---|
| Subject | Who or what is on screen | A young chef in a white apron |
| Action | One clear action | Flips vegetables in a flaming wok |
| Setting | Place and time | Busy restaurant kitchen at night |
| Camera | Framing and movement | Close-up, slow push-in, shallow focus |
| Lighting | Light source and quality | Warm tungsten light, flames, steam |
| Style and mood | Look and feeling | Cinematic, energetic, natural colour |
| Sound | If supported | Sizzling, kitchen chatter, no music |
Which camera words work best?
- Framing: Wide shot, medium shot, close-up, extreme close-up, over-the-shoulder.
- Movement: Static, pan, tilt, dolly in or out, tracking, orbit, crane, handheld.
- Angle: Eye level, low angle, high angle, overhead, Dutch angle.
- Lens and focus: Wide-angle, telephoto, shallow depth of field, rack focus.
- Speed: Slow motion, real time, time-lapse.
How do you describe lighting and style?
- Name the light source: window light, neon signs, candlelight, overcast sky
- Name the time of day: golden hour, blue hour, midday sun, night
- Describe quality: soft, hard, high contrast, low key
- Describe colour: warm, cool, desaturated, vibrant, teal and orange
- Reference a format rather than a living artist: 35mm film, documentary, music video, product commercial
How do you build a consistent sequence?
- Write a shot list: Four to eight shots: establishing, medium, close-up, detail, hero.
- Fix a style line: One sentence of style and colour words reused in every prompt.
- Use references: The same character, product or location image for every shot.
- Prompt each shot separately: Keep one action per prompt.
- Generate two to four takes: Pick the best of each.
- Edit together: Assemble, add music, titles and logo in your editor.
How do you fix a weak result?
Change one thing at a time so you know what made the difference, and keep notes of prompts that work for your brand.
What does a full example look like?
A restaurant wants a 20-second reel. The shot list has four shots, all sharing the style line “warm, cinematic, natural colour, shallow depth of field”. Shot one: wide establishing shot of a busy open kitchen at night, slow dolly in. Shot two: close-up of a chef flipping vegetables in a flaming wok, slow motion, sizzling sound. Shot three: overhead shot of plating, hands placing herbs on a dish. Shot four: hero shot of the finished dish on a dark table, steam rising, gentle orbit. Each shot gets three takes; the editor picks the best, adds the logo and music, and the reel is ready in an afternoon.
How should you prompt for sound?
Some models, such as Veo, generate audio with the picture; others produce silent clips. Where sound is supported, describe it as precisely as the visuals: the main sound, the background and whether there is music or dialogue. “Sizzling wok, distant kitchen chatter, no music” works better than “restaurant sounds”. For dialogue, keep lines short and put them in quotation marks. In most brand work, add music and voice-over in the edit, where you can control licensing, timing and levels, and use generated sound for ambience.
Should you keep a prompt library?
Yes. Save every prompt that produced a usable shot, with the tool, settings, reference images and a thumbnail. Group them by shot type and brand. Over time the library becomes a house style that new team members and clients can follow, and it makes results far more predictable from one project to the next.
What mistakes should you avoid?
- Describing several actions or locations in one prompt
- Changing style words between shots
- Relying on AI for logos and on-screen text
- Using real people’s likenesses without consent
- Skipping the edit; AI clips need assembly and sound
How UIUX Studio helps
We plan and produce AI-assisted video for concepts and brand content. For reels and social video at volume, Social Mafia by UIUX Studio handles editing and AI-assisted production. Our product design team supports brand visuals.
Frequently asked questions
What makes a good AI video prompt?
A single shot described with subject, action, setting, camera, lighting, style and sound.
Do prompts work the same in every tool?
The structure works broadly; each tool has its own strengths and limits.
How long should a prompt be?
One to three sentences per shot is usually enough.
How do I keep characters consistent?
Use the same reference image and style words in every shot.
Should I name famous directors or artists?
Describe the look instead, such as lighting, colour and format.