Gemini Omni Prompt Guide

Start With The Shot

Gemini Omni prompts work best when they read like a compact shot brief. Name the subject, what changes during the clip, how the camera moves, what references matter, and what the final viewer should hear or read.

Subject: a ceramic perfume bottle on a wet stone surface
Action: the bottle rotates as mist rolls across the table
Camera: one continuous macro push-in, shallow depth of field
Lighting: soft studio key light, colored rim light, realistic reflections
Sound: subtle glass tap, airy room tone
Format: 9:16, 10 seconds, product launch ad
Constraints: no extra text, keep the label readable, no scene cuts

Prompt Anatomy

Change The Action

Ask for the movement you want, not only the visual style. Good action prompts describe timing and cause: an object opens, a surface ripples, a character reacts, or the environment changes in sync with music.

Keep the room, subject, and lighting the same.
When the person touches the mirror, make the mirror ripple like liquid.
The arm becomes reflective material for two seconds, then returns to normal.
Use one continuous shot. No cuts.

Use simple verbs first: enters, turns, opens, ripples, dissolves, falls, reacts, follows, transforms. If the action is already clear, do not over-describe every frame.

Apply World Knowledge

Gemini Omni is positioned around Gemini's world knowledge, so prompts for physics, history, science, and explainers can be shorter than a fully scripted animation brief. Write the rule that matters, then let the scene breathe.

Claymation explainer of protein folding.
Everything is made of clay, no hands visible, stop-motion pacing.
Show the chain folding into a stable shape because hydrophobic regions move inward.
Accurate enough for a high-school biology class.

For explainers, include:

• the phenomenon, such as gravity, refraction, kinetic energy, folding, orbit, or fluid motion;
• what must remain accurate;
• whether labels, lower thirds, or diagrams should appear;
• the viewer level, such as beginner, product demo, classroom, or executive summary.

Render Text Deliberately

Text is part of the shot. If you need readable words, specify the exact words, placement, duration, style, and motion.

Word by word, one word on screen at a time:
"did", "you", "know", "this", "model", "can", "render", "text".
Each word uses a different animated style.
Paced to a fast sizzle-reel rhythm, centered, high contrast, no misspellings.

Avoid asking for long paragraphs inside video. Use short labels, title cards, lower thirds, or one-word beats.

Direct The Camera

Camera language should be specific enough for the model to choose a coherent shot.

Reference With Real Controls

Google's prompt guide shows the broader Omni model family combining video, image, audio, and text references. GemiOmni's current Seedance-backed workspace is narrower: use text prompts plus up to two image references, then describe motion, sound intent, and storyboard beats in the prompt instead of uploading separate video or audio references.

Apply New Styles

Style prompts should say what changes and what remains fixed.

Create a four-part style progression of the same walk cycle:
1. vibrant crayon texture on granulated paper;
2. graphite pencil sketch with cross-hatching;
3. claymation with handmade surface texture;
4. cinematic realism with the same camera path.
Keep the subject, action, and frame center consistent.

Add A Storyboard

If you already know the story, write it as beats instead of a single dense paragraph.

Entire story in 10 seconds, cinematic, 16:9.
Beat 1: a small paper boat floats under a desk lamp.
Beat 2: the lamp becomes a moon over a paper sea.
Beat 3: the boat passes folded paper mountains.
Beat 4: final wide shot, all elements visible, no text.

Iterate Without Losing The Thread

Follow-up prompts should change one or two variables while preserving the rest.

Keep the same subject, camera path, product label, and lighting direction.
Change only the environment: make it a black-and-white checkerboard studio.
Add soft reflections on the floor.
Do not change the bottle shape.

Useful edit targets:

• replace or restyle one object;
• transform the environment;
• add motion effects;
• change the camera angle;
• shift style to anime, claymation, watercolor, documentary, or cinematic realism;
• preserve the original subject while changing the background.

GemiOmni Checklist

Before generating, check:

• Aspect ratio: choose the final channel first, usually 16:9 or 9:16.
• Duration: short clips need one clear idea, not a whole commercial.
• References: each upload should have a job.
• Sound: only ask for sound when the current model path supports it.
• History: preserve prompt, model, settings, references, and output URL so the next retry can be controlled.

Sources

• Google: Introducing Gemini Omni
• Google DeepMind: Gemini Omni
• Google DeepMind: How to create effective prompts with Gemini Omni