Multilingual
Visit website
Updated: 2026-09-30 08:18
Access: Normal
Profile views 3 
Outbound visits 0 Website Review
What is Gemini Omni Prompts?
Gemini Omni Prompts is an independent, unofficial prompt library for Google's Gemini Omni video model. It collects copy-paste-ready prompts adapted from Google's official prompt guidance and from community testing, and it labels each prompt with its source so you can see where the pattern came from. It is not affiliated with Google.
Its structure is built for people who want to produce video rather than read theory:
- Prompt library — around 20 hand-picked prompts across 9 categories, with featured examples such as a cinematic continuous shot, an aerial pull-back over a surfer, a slow-motion vase shatter, and a mirror-ripple VFX transformation.
- Prompt guide — explains the recurring techniques the prompts rely on, such as naming specific camera-motion verbs instead of generic "drone shot," stating duration and aspect ratio in the opening line, and using "when X happens, Y changes" trigger syntax for effects.
- Video walkthroughs — weekly source-cited walkthroughs showing the prompt, the actual Gemini Omni output, and commentary on what holds up.
- Language coverage — prompts are offered in 5 languages.
The useful distinction is between the prompts themselves and the patterns behind them. A prompt like the vertical cosmetics splash is a finished asset for a Reels-style beauty ad; the guide is what lets you adapt the same structure to your own product, shot length, or format.
If you are evaluating it, start with one featured prompt in your target aspect ratio, run it, then change a single variable — camera verb, duration, or trigger phrase — and compare outputs. That tells you quickly whether the library's patterns transfer to your subject. Two related references are Google DeepMind for the official prompt guidance the library cites, and Google AI for model-level documentation.
How do I adapt a prompt from this library for my own video without copying it verbatim?
Adapt by keeping the structural pattern and swapping the subject matter — that's the real value of a library like Gemini Omni Prompts. The featured entries show reusable skeletons: an opening line that fixes shot grammar ("one continuous shot", duration, aspect ratio), a camera-motion verb, and a physics or VFX trigger. Those are the parts worth borrowing; the rainy window, the surfer and the vase are not.
A three-step swap
- Copy the skeleton, not the sentence. From the "Cinematic Continuous Shot" entry, the transferable pattern is [shot type] + [duration] + [aspect ratio], then [subject action], then [camera verb]. Rewrite it around your own subject.
- Replace every concrete noun. "Designer's rainy window" becomes your location and your character. Keep the abstraction level identical — if the original says "controlled liquid splash", don't write "some liquid moving".
- Keep the trigger clause intact. The "when X happens, Y changes" syntax is the mechanism, not decoration. Change X and Y to your objects; leave the grammar alone.
What to change and what to leave
| Element | Keep | Replace |
|---|---|---|
| Shot grammar (duration, aspect ratio) | Yes | Only if your platform differs |
| Camera-motion verbs ("pull-back", "push-in") | Yes | — |
| Trigger syntax ("when X, Y changes") | Yes | The X and Y themselves |
| Subject, setting, props | — | Always |
| Stylistic adjectives | — | Match your own reference |
A concrete case
Say you shoot a 9:16 vertical for Reels. The cosmetics splash prompt is the closest template: vertical framing plus a slow-motion constraint to keep fluid simulation coherent. If your product is a coffee pour, keep "9:16", keep the slow-motion constraint, keep the hero-shot framing, and replace the serum bottle and splash with your cup and pour. The physics advice transfers; the beauty-commercial styling does not.
A practical check
Run your version and compare it against the library entry side by side. If your output drifts — wrong pacing, broken physics, ignored aspect ratio — you probably changed a structural element rather than just the subject. Revert that one piece.
Two cautions. The site describes itself as independent and unofficial, and the entries are adapted from Google's prompt guide and community testing rather than being Google's own wording — so treat the patterns as informed convention, not specification. Also, the page shows a "subscribe" signal but no pricing detail, so check what's actually free before relying on it as a long-term reference.
For a second opinion on the same technique, Google's own prompt guidance is the primary source to compare against.
Which prompts are best for testing Gemini Omni's physics engine with realistic motion?
For physics testing, pick prompts that isolate one physical behavior and let the model show it clearly. The best candidates from this library are the Slow Motion Physics Test: Ceramic Vase Shatter (verified pattern, 16:9, 10s) and the Cosmetics Liquid Splash commercial shot, which is built around a slow-motion constraint that keeps fluid simulations coherent. Both give you a visible failure mode: fragments that float or fluid that behaves like jelly.
What makes a good physics test prompt
- One phenomenon per prompt. Shatter, splash, or drag — not all three.
- A slow-motion or high-frame-rate cue. Fast action hides errors; slow motion exposes them.
- A clear trigger. "When X happens, Y changes" syntax lets you check whether the model respects cause and effect.
- Fixed camera and duration. A moving camera makes it hard to tell whether the physics or the motion is wrong.
A practical test loop
Run the vase shatter first as your baseline, then re-run it with one variable changed (duration, aspect ratio, or added lighting). The library's "Watch each prompt tested live" section shows the actual outputs, which is useful for comparing your result against a known-good run before you conclude the model failed.
If you want a second opinion on physics handling in a different model, Google DeepMind publishes official prompt guidance, and the community-verified notes here are adapted from it.
How can I tell whether a prompt here is backed by official Google documentation versus community testing?
The site labels each prompt with a source tier, and that label is the main thing to read. From the page evidence, prompts are described as "adapted from Google's official documentation and verified community testing," and individual entries carry tags like "VERIFIED" alongside a stated origin — for example, one prompt is credited to "Google's DeepMind prompt guide," another to "PixVerse's hands-on Gemini Omni test," and another to "Medium's Gemini Omni Prompt Playbook." So the practical distinction is: official-doc prompts trace to Google/DeepMind guidance, while community prompts trace to a named tester or publication.
H3 How to read the difference in practice
- Official-doc origin: the prompt is adapted from Google's own prompt guide or a DeepMind demo. Expect general, durable patterns — camera-motion vocabulary, trigger syntax like "when X happens, Y changes," and structural conventions.
- Community-verified origin: the prompt was tested by a named creator or outlet and reported to work. Expect more specific recipes — exact durations, aspect ratios, and shot setups — but also more dependence on one person's setup and one model version.
The page also mentions "source transparency tiers," which suggests the site itself ranks how directly a prompt connects to its source. If you need certainty, open the cited source for the prompt you care about and check whether it says what the prompt claims.
H3 A quick decision rule
If you are learning the model, start with officially sourced prompts — they teach reusable structure. If you need a specific shot today, a community-verified prompt is often more concrete, but treat it as a starting point and re-test it yourself, since results can shift with model updates.
For comparison, Google's own guidance lives at Google AI for Developers, and DeepMind publishes research and demos at Google DeepMind.
What camera-motion phrases should I use to get a specific effect instead of a generic drone shot?
Use motion verbs that describe the camera's path and speed, not the equipment. "Drone shot" tells the model almost nothing; "pull-back," "push-in," "rotate revealing," or "flythrough" tell it where the camera starts, where it moves, and what it uncovers.
Motion vocabulary that maps to effects
- Pull-back / pull-out — camera retreats, widening the frame to reveal context around a subject.
- Push-in — camera advances toward a subject; builds intimacy or tension.
- Rotate revealing — camera arcs or turns, bringing a new element into view.
- Flythrough — camera travels forward through a space, usually tight or enclosed.
- FPV dive — fast, steep descent with a first-person feel.
- Continuous shot — one unbroken take; pair with a duration so the model knows the length.
- Aerial reveal — start close or partial, then rise or widen to show scale.
How to combine them
Gemini Omni Prompts' featured examples show the pattern: put the motion verb plus duration and aspect ratio in the opening line, then add the scene. Their "Aerial Reveal: Pull-Back Over Lone Surfer at Golden Hour" uses "pull-back" and "rotate revealing" rather than "drone shot," and their slot-canyon prompt leans on reveal and push vocabulary for an FPV flythrough. The library also flags that FPV prompts often fall into brand-name traps, so describe the movement and the space instead of naming a camera or drone model.
A concrete reader scenario
If you want a product hero shot, "slow push-in on a serum bottle, then pull-back to reveal the full counter" gives you two distinct beats. If you want landscape scale, "pull-back from a lone surfer to reveal the coastline at golden hour" gives the model a start point and an end point. Add "one continuous shot, 10 seconds, 16:9" when you need an unbroken take.
Next step
Write your prompt as start position → motion verb → end reveal. Then check the result against the effect you named; if the camera moved the wrong way, swap the verb before adding more scene detail. For more worked examples, browse Gemini Omni Prompts.
How do I format a vertical prompt for Reels or Shorts rather than a widescreen video?
Put the aspect ratio and duration in the opening line of the prompt — not at the end — and pair them with a vertical-native subject framing. On Gemini Omni Prompts, the library's own featured examples follow this pattern: the cinematic and drone prompts lead with 16:9 10s, while the beauty commercial example is tagged 9:16 7s and is described as "framed vertically for Reels."
What to change versus a widescreen prompt
- Ratio first. Start with
9:16before any scene description. Models tend to weight the earliest tokens, and burying the ratio after a long shot description is the most common reason you get a letterboxed result. - Duration matches the platform. Shorts and Reels reward tight loops. The library's vertical example uses 7s; its widescreen ones use 10s. Shorter clips also mean less chance of the composition drifting out of frame.
- Reframe the subject, don't crop it. A vertical frame is tall and narrow, so a wide landscape, a crowd spread across the horizon, or a lateral tracking shot will lose its subject. Choose vertical-friendly subjects: a standing figure, a bottle on a counter, a face, a vertical wall, a descent or climb through a narrow space.
- Change the camera verb. Horizontal motion verbs (pull-back, pan, flythrough) fight the frame. Vertical-friendly alternatives: push-in, tilt-up reveal, top-down, overhead-to-face, rising crane.
- State safe zones if it matters. If you will overlay captions or UI, describe the subject as occupying the centre band and leaving headroom and footroom clear.
A practical template
9:16, 7s — [subject] in [setting], one continuous shot,
slow push-in from [start framing] to [end framing],
[lighting], [motion detail], no camera pan.
The library's own "one continuous shot" example is the useful model here: it puts ratio, duration and shot type in the opening line, then spends the rest of the prompt on content. Do the same, just swap 16:9 for 9:16 and replace any lateral camera movement.
A concrete scenario
You want a 6-second Short of a serum bottle. A widescreen version might read "wide shot, bottle on a marble counter, camera pans right." The vertical rewrite: 9:16, 6s — serum bottle centred on a dark counter, one continuous shot, slow push-in, controlled liquid splash, soft rim light, no pan. Same subject, but the framing and motion now work in a tall frame.
Decision criterion
If the shot's meaning depends on horizontal space — a landscape, a chase across a room, two people side by side — don't convert it; storyboard a different shot for vertical. If the meaning depends on the subject's height or on depth toward the camera, the same idea converts cleanly with a ratio swap and a camera-verb change.
Next step
Generate the same prompt at 9:16 and 16:9 back to back and compare. If the vertical version crops the subject's head or loses the product, the fix is almost always the framing sentence, not the ratio tag.
Hosting and Email
Cloudflare
Hosting
Location
Pages, Search and Sharing
All bots
Registration details 
Identified technologies
User reviews (0)