Video

Best Gemini Omni Prompts for AI Video: Camera, Motion, Style, and References

By Sagnik Bhattacharya 20 May 2026 8 min read

Coding Liquids blog cover featuring Sagnik Bhattacharya for Gemini Omni prompts, with camera direction cards, motion arrows, style swatches, and reference assets.

Good Gemini Omni prompts are less about piling on adjectives and more about controlling the shot. The official prompt guide points creators towards framing, motion, style, lighting, location, action, and references. This post turns those ideas into practical prompt blocks you can reuse.

I teach Flutter and Excel with AI — explore my courses if you want structured learning.

If you have not made a first clip yet, start with the beginner Gemini Omni tutorial. If you already have a clip and want to change it, pair this with the multi-turn editing guide.

Note: Prompt examples here are written as templates. They are not claims that every model run will obey every instruction perfectly.
Note: For fair testing, hold duration, aspect ratio, reference assets, and active model constant before comparing prompt variations.

Quick answer

The best Gemini Omni prompt starts with the subject and action, then adds framing, camera motion, lighting, style, references, constraints, and review targets. Put the most important visual requirement early, and use follow-up prompts for edits.

You need camera and motion language that the model can interpret.
You want prompt templates for product videos, explainers, Shorts, edits, and image-to-video scenes.
You want a repeatable prompt test method instead of random regeneration.

The Prompt Anatomy That Works Best

A strong Gemini Omni prompt reads like a tiny director's brief. It does not have to be long, but it must be ordered. Start with what the viewer should see. Then explain how the camera behaves. Then add atmosphere, references, and constraints.

The table below is the structure I would use for nearly every first draft. You can shorten it once you know which parts matter for your project.

Prompt part	What to write	Example
Subject	The main person, object, place, or chart.	A silver laptop on a clean desk
Action	One visible behaviour or transformation.	The screen opens and a dashboard animates in
Framing	Wide, medium, close-up, overhead, over-the-shoulder.	Medium shot, slightly above desk level
Camera	Static, push in, dolly, pan, tilt, handheld, locked-off.	Slow push in, one continuous shot
Lighting	Source, direction, colour temperature, contrast.	Soft morning light from the left
Style	Documentary, product demo, cinematic, editorial, watercolour.	Clean SaaS product-demo style
Constraints	No text, no extra people, preserve reference, simple background.	No on-screen text; keep the desk uncluttered

Camera Prompts

Camera language is where many AI video prompts become better immediately. Instead of asking for a generic cinematic clip, name the camera relationship to the subject. Gemini Omni's prompt guide explicitly suggests framing and camera motion, including ideas such as fixed, push in, dolly zoom, and natural smartphone zoom.

Use one camera move per prompt until the model behaves. If you need several moves, make them sequential and keep the scene simple.

Locked product shot: A locked-off medium shot of a smart watch on a stone surface. The watch face glows softly. No camera movement.
Gentle reveal: Start in a close-up on a closed notebook, then slowly tilt up to reveal a tidy planning board behind it.
Over-the-shoulder tutorial: Over-the-shoulder view of a designer arranging mood board cards on a tablet. One continuous shot, slow handheld movement.
Dolly zoom warning: Use dolly zoom only when the scene is simple; it can look dramatic but often exposes motion instability.

Motion Prompts

Motion should describe what changes over time. A still visual prompt plus the word cinematic is not enough. Name the motion of the subject, the environment, and the camera separately if they all matter.

The safest beginner pattern is one subject motion plus one camera motion. More than that can still work, but it becomes harder to diagnose failures.

Create an 8-second landscape video of a glass coffee cup on a cafe table. Steam rises slowly from the cup while the camera makes a very slow push in. Afternoon window light, realistic documentary style. Keep the background softly blurred and do not add text.

Style And Lighting Prompts

Style is not just a colour palette. It affects texture, camera behaviour, animation rhythm, and how realistic the scene should feel. The prompt guide describes realistic, cinematic, grounded, majestic, and other style directions; treat those as starting words, then make them concrete.

A reliable style prompt says what to preserve. For style transfer, specify whether the original motion, composition, character, or object identity should remain intact.

Apply a clean studio product-ad style while preserving the same product shape and camera path.
Reimagine this clip as warm watercolour animation, preserving the person's pose and the direction of movement.
Make the lighting crisp and high contrast, with a single key light from the right and a dark neutral background.
Use a flat editorial explainer style with simple geometric shapes, but keep the original timing and scene order.

Reference Image And Storyboard Prompts

References are the difference between describing a world and giving the model a target. Gemini Omni can use images, text, video, and audio as references, and the prompt guide highlights references for consistency and storyboards for narrative order.

When you use a reference, say which part of the reference matters. Otherwise the model may preserve the wrong thing.

Reference type	Prompt instruction	Best use
Product image	Use the bottle shape and label from the image; animate only the camera.	Product demos and ads
Character sheet	Keep the same face, outfit, and proportions across the whole clip.	Recurring characters
Storyboard	Follow the panels in order from top left to bottom right.	Explainers and Shorts
Audio	Sync the light pulses to the beat, but keep the camera locked.	Music-led visuals

Prompt Testing Framework

To benchmark prompts legitimately, test prompt variants rather than relying on vibes. Use the same subject, duration, aspect ratio, model, and reference assets. Change only one prompt variable at a time: camera wording, style wording, constraint wording, or reference wording.

Score every output on prompt adherence, motion stability, text rendering, audio usefulness, character or object consistency, editability, and cost per usable clip. The winning prompt is not always the prettiest one; it is the one that produces the most usable clips per batch.

Prompt Templates You Can Adapt

These templates are intentionally specific. Replace the bracketed parts, then keep the structure. If a result fails, fix the part that failed rather than rewriting everything.

Product demo: Create a [duration] [aspect] video of [product] in [setting]. [Action]. Camera: [framing and movement]. Lighting: [source and mood]. Style: [visual style]. Constraints: [what not to include].
Explainer: Visualise [concept] using [style]. Show [three visual beats] in order. Keep text minimal and readable. Use [motion rhythm].
Reference edit: Keep [reference element] exactly the same. Change only [one element]. Preserve [motion/composition/lighting].
Shorts hook: Open with [visible surprise] in the first second. Vertical 9:16. One fast but readable camera move. No tiny text.

Step-by-step: build one production prompt

The easiest way to write better Gemini Omni prompts is to build them in layers. Each layer should add control, not decoration. Use this process when a prompt needs to be reused for clients, Shorts, product clips, or spreadsheet explainers.

Start with the output job. Write what the clip must do: "show the product opening", "explain the sales increase", or "turn this sketch into realistic footage".
Add the subject and action. Name the object, character, or reference and describe only the main movement. If there are three actions, split them into three clips.
Add framing. Choose close-up, medium shot, wide shot, vertical, or landscape. If the subject must stay readable on a phone, say so.
Add camera behaviour. Use one camera instruction: locked-off, slow push-in, over-the-shoulder, dolly left, or one continuous shot. The official prompt guide recommends specific videography directions such as fixed, push in, punch in, or dolly zoom.
Add lighting and style. Use a style that helps the goal: clean product demo, documentary, claymation explainer, flat-media explainer, or realistic smartphone footage.
Add preservation rules. For references, list exactly what should stay the same: shape, logo position, chart trend, outfit, room layout, or motion path.
Add failure prevention. Tell the model not to add text, extra objects, new people, brand changes, or misleading numbers when those would break the result.
Add the review target. End with what the final frame should support: captions, a loop, a cut to another clip, or a product close-up.

Here is the same prompt growing from weak to useful:

Weak: Make a cool video of my dashboard.

Useful: Use the uploaded dashboard screenshot as a reference for an 8-second landscape business explainer. Preserve the upward revenue trend and the ranking of regions. Camera slowly pushes in from the full dashboard to the revenue chart. Clean corporate motion-graphics style, blue and green accents, no generated numbers or tiny text. End on a stable final frame with space for captions added later.

Prompt QA before you generate

Run this quick quality check before spending credits. If a prompt fails two or more rows, rewrite it before generating.

Question	Good answer	Weak answer
What is the clip for?	One clear job.	"Make it cinematic."
What should move?	One main action.	Several unrelated actions.
How should the camera behave?	One clear camera move.	No camera direction or too many moves.
What must stay consistent?	Specific reference details.	"Keep it the same" with no detail.
What should be avoided?	Concrete failure limits.	No constraints.

Common mistakes

Using five style adjectives but no action.
Changing the prompt and the reference at the same time during testing.
Requesting perfect typography inside the generated clip when editing captions later would be cleaner.
Forgetting aspect ratio until after generation.
Treating a good prompt as permanent even after the model or Flow interface changes.

Sources

These official references are useful if you need the product or framework documentation alongside this guide.

Want to create better AI content?

My courses cover practical AI workflows for content creation, video production, and marketing with real projects.

Browse courses