Gemini Omni Flash on PixVerse: A Step-by-Step Prompt Workflow

Gemini Omni Flash is one of the video model options available on PixVerse, and it rewards a bit of planning before you hit generate. This tutorial walks through a repeatable workflow: choose your input type, lock in duration and aspect ratio, build a structured prompt, and review the output like a production asset rather than a lucky roll.

You can run this same workflow for text-to-video, image-to-video, or multi-image reference generation, then compare the result against other models in the PixVerse workspace, including PixVerse V6 and PixVerse C1, before deciding which clip to ship.

Before You Start: What This Workflow Covers

PixVerse’s current Gemini Omni Flash integration is a generation workflow, not the full conversational editing surface Google documents for its API. Knowing the boundaries up front saves you from planning a shot list the model can’t yet deliver on PixVerse.

Capability	Available on PixVerse today	Planning note
Text-to-video	Yes	Good for original concepts and quick creative tests.
Image-to-video	Yes	Good for turning a still (product, poster, sketch) into motion.
Reference image-to-video	Yes, up to 5 images	Reference each as `@image1` through `@image5` in your prompt.
Clip length	3-10 seconds	Pick the shortest length that still tells the idea clearly.
Output	720p, 16:9 or 9:16	Match the ratio to the destination channel before you generate.
Audio	Prompt-driven, synced	You direct ambience, effects, or silence through the text prompt.
Natural-language editing, extension, transitions, video/voice reference	Not yet on PixVerse	Use another PixVerse model when these controls are the priority.

Step 1: Pick Gemini Omni Flash as Your Model

From the PixVerse generation screen, open the model picker and select Gemini Omni Flash if it’s available on your plan. Because PixVerse hosts multiple models side by side, you can run the identical brief through Gemini Omni Flash and then re-run it on another model to see which one nails your specific shot.

If you don’t see Gemini Omni Flash yet, check your account tier and the in-app rollout notice — availability and credit cost can shift, so treat the model picker as the current source of truth rather than this guide.

Step 2: Decide Your Starting Point

Your prompt strategy depends entirely on what you’re feeding the model:

Text-to-video — best when there’s no existing asset to build from: new concepts, social clips, quick ad drafts, explainer visuals.
Image-to-video — best when you already have a still that should anchor the scene: a product photo, key art, a poster, packaging.
Reference image-to-video — best when several images need to jointly steer the result. Upload up to five JPEG/PNG files and call them out individually in your prompt (@image1, @image2, etc.), telling the model exactly what each one controls — subject, palette, lighting, pose.

Step 3: Lock In Duration and Aspect Ratio First

Set these before you write the prompt, since they shape how much story the prompt needs to carry.

Duration: 5-6 seconds is enough for a single product beat; 8-10 seconds gives room for a beginning, middle, and end (character intros, explainers).
Aspect ratio: 9:16 for TikTok, Reels, and Shorts; 16:9 for YouTube, landing pages, and decks. Generate each ratio as its own job rather than cropping one master clip to fit multiple placements.

Step 4: Build the Prompt Like a Mini Production Brief

A prompt that only describes a vibe (“make it look premium”) gives the model far less to work with than one that specifies the scene. Work through these building blocks in order:

Subject — the person, product, or object anchoring the frame.
Action — what physically changes over the clip.
Camera — close-up, wide, push-in, orbit, handheld, overhead, locked-off.
Lighting and mood — daylight, neon, soft studio, dramatic, playful.
Environment — location, props, weather, materials, texture.
Audio — ambience, effects, music mood, or explicitly “no dialogue.”
Timing — what happens in which seconds, for anything longer than 6-7 seconds.
Constraints — what to avoid: logos, extra text, scene cuts, celebrity likeness.

A simple template to fill in:

Create a [duration] [aspect ratio] video. The subject is [specific subject]. The action is [specific motion]. The camera [movement and framing]. The environment is [place, lighting, materials, props]. Audio: [ambience, effects, music mood, or silence]. Constraints: [what to preserve, what to avoid].

Force a Single Continuous Shot When You Need One

Left unguided, Gemini Omni Flash may cut between several mini-shots inside one clip. For product shots, fashion motion, or anything where a cut would break continuity, add “single continuous shot,” “no scene cuts,” or “one unbroken take” directly into the prompt.

Write Audio Direction Into the Prompt, Every Time

Since audio is generated in sync with the video, don’t leave it to chance. Describe the ambience or effects you want (soft clicks, room tone, crowd energy), and say “no dialogue” if the clip should stay wordless. Avoid asking for a specific song, artist, or known voice style — describe the mood instead.

Add Timecodes for Multi-Beat Clips

For 8-10 second clips with more than one story beat, timecoded instructions keep the model anchored to your structure:

[0-2s] Establish the product on the table. [2-5s] Camera pushes in as light catches the surface. [5-8s] The product rotates slightly and settles on a clean end frame.

Keep it to two or three beats — cramming more into a 10-second window tends to blur the most important action.

Tell the Model What Each Reference Image Does

If you’re using more than one reference image, ambiguity is the enemy. Spell out the job of each one:

Use @image1 as the exact product reference — preserve the bottle shape, cap color, and label placement. Use @image2 only for lighting and background mood. Do not carry over any logos or people from @image2.

Step 5: Generate, Then Review Like a Producer, Not a Fan

Once the clip renders, judge it against the job it needs to do, not just how it looks at first glance. Check:

Did the subject/product stay visually consistent throughout?
Is any on-screen text readable, or garbled?
Does the audio actually match the motion on screen?
Is the final frame usable as a thumbnail or cut point?

For commercial use, also screen for rights issues — avoid anything that leans on a real person’s likeness, a branded logo, or a recognizable song or voice. For a deeper look at how Gemini Omni Flash handles safety and watermarking, see our Gemini Omni Flash overview.

Five Prompt Recipes to Start From

Use these as starting templates, then swap in your own product, subject, and constraints.

Recipe 1: Product Photo to Vertical Ad

Setup: 9:16, 6-8 seconds, image-to-video or one product reference image.

Create an 8-second 9:16 product video using @image1 as the exact product reference. Preserve the product shape, cap, color, label placement, and silhouette. The product stands on a wet dark stone surface at sunrise. [0-2s] Macro close-up of water droplets on the surface. [2-5s] Camera slowly pulls back as warm light catches the edges. [5-8s] The product rotates slightly and settles centered. Audio: soft water droplets, subtle room tone, no dialogue. Constraints: no extra logos, no extra text, no scene cuts.

Recipe 2: Educational Explainer

Setup: 16:9 or 9:16, 8-10 seconds, text-to-video.

Create a 10-second 16:9 explainer about how a solar panel converts sunlight into electricity, using a paper-craft style on a dark tabletop. [0-3s] A paper sun sends rays toward a blue panel. [3-6s] Glowing dots travel along a drawn circuit. [6-10s] A small paper house lights up. Camera: locked-off overhead shot with small stop-motion movements. Text: only “sunlight”, “panel”, “electricity”, spelled exactly. Audio: soft paper movement, a tiny chime, no voiceover. Constraints: no extra words, no human hands, no logos.

Recipe 3: Character Introduction from References

Setup: 16:9, 8-10 seconds, up to three reference images.

Create a 9-second 16:9 original character intro. Use @image1 as the identity reference — preserve face shape, hairstyle, jacket color, silhouette. Use @image2 only for lighting and city-background mood. The character stands on a rooftop at dusk, turns toward camera, lifts a small glowing device. Camera: slow medium push-in, single continuous shot. Lighting: blue evening sky, warm rim light. Audio: distant city ambience, a soft electronic hum. Constraints: original character only, no franchise references, no logos, no dialogue.

Recipe 4: App Hero Loop

Setup: 16:9, 6-8 seconds, text-to-video.

Create a 7-second 16:9 hero loop for an AI planning app. A translucent floating timeline hovers above a clean desk, with abstract cards and lines arranging into a calm weekly plan. Camera: slow left-to-right slider, shallow depth of field, single continuous shot. Lighting: natural morning light, soft shadows, teal and warm-yellow accents. Audio: subtle interface clicks, soft ambient tone. Text: no readable app name or task text. End frame should match the opening frame so the clip loops cleanly.

Recipe 5: Fashion Lookbook Motion

Setup: 9:16, 8-10 seconds, reference image-to-video.

Create a 10-second 9:16 fashion lookbook clip. Use @image1 as the outfit reference — preserve coat length, fabric texture, palette, shoe style. Use @image2 for studio lighting only. A fictional model walks across a concrete studio, pauses, turns one shoulder to camera as the coat moves naturally. Camera: vertical full-body framing, smooth dolly, no scene cuts. Lighting: large softbox from the left. Audio: quiet studio ambience, soft footsteps, no dialogue. Constraints: fictional model, no celebrity likeness, no brand logos.

FAQ

Do I need Gemini Omni Flash specifically, or can I use another PixVerse model?

Gemini Omni Flash is one option among several on PixVerse. It’s a strong fit for prompt-driven scenes with synced audio and multi-image reference control; other PixVerse models may suit extension, transitions, or longer-form edits better until those land for Gemini Omni Flash.

How many reference images can I upload?

Up to five JPEG or PNG images per generation, referenced individually in your prompt as @image1 through @image5.

What resolution and length should I expect?

The current PixVerse integration outputs 720p video, 3-10 seconds long, in 16:9 or 9:16.

Can I get exact on-screen text in a clip?

Keep on-screen text short and explicit in the prompt (exact words, quoted), and expect to review it closely — precise typography is still an area where results can vary.

Is video extension or transition supported yet?

Not in the current PixVerse Gemini Omni Flash workflow. Use another PixVerse model when extension, transition, or video/voice reference is the priority.

Ready to try it yourself? Generate your first Gemini Omni Flash video on PixVerse.