The short version. Google AI published a Gemini Omni prompting guide on May 26, 2026 covering five techniques for getting better video output. This post unpacks each one with the original prompts, real example outputs, and three bonus tricks the community has surfaced (style transfer, audio translation, environment swap).
Where you can use Gemini Omni Flash today. The Gemini app, Flow by Google, Google Flow Music, plus YouTube Shorts and Create. Free for most use cases.
Stop describing the world to the model. Use the cultural shorthand. “Astronaut’s POV on Mars” works because Omni already knows what Mars looks like, what an astronaut helmet does to the field of view, and how the dust kicks up. The skill is knowing what you can skip.
Gemini Omni is Google’s newest video-capable model. The Google AI team published a hands-on prompting guide on May 26, 2026 covering five techniques that matter for quality video output. The community immediately produced demos extending the techniques in directions Google did not list: cross-language audio translation, environment swaps, style transfers across whole shots. This post collects both.
Every prompt below comes from the original guide or from verified community demos. Every screenshot is from a real Gemini Omni Flash output. If you want to follow along, open the Gemini app, switch to video generation, and try the prompts as you read. For a broader set of prompt patterns across every model, the Best AI Prompts by Job and Use Case hub is the right starting point.
What is Gemini Omni and how do I access it?
Gemini Omni is Google’s multimodal generation model. The “Omni” name signals the goal: create anything from any input, starting with video. The first widely-available tier is Gemini Omni Flash, the faster lower-cost variant. The full Omni model is still rolling out.
- Gemini app. Web and mobile. Free tier includes Omni Flash. Best entry point for beginners.
- Flow by Google. Google’s filmmaking surface. Designed for longer sequences and shot-by-shot editing.
- Google Flow Music. Audio-side companion. Useful when you want music synced to video output.
- YouTube Shorts and Create. The mobile creator surface. Generate Shorts directly inside the YouTube app.
The same model behaves slightly differently across surfaces because each one wraps the generation API with its own UI controls. For the Google AI team’s five core techniques, the Gemini app is the cleanest place to experiment.
How do I tap Gemini’s real-world knowledge?
Technique 1. Skip the granular descriptions. Omni was built on the same training that makes Gemini understand history, science, and culture, so it can render realistic-looking content from cultural shorthand. Use historical eras, scientific terms, or cultural touchstones directly. Stop over-explaining.
Real prompts from the Google AI thread:
- “Astronaut’s POV on Mars”
- “A marble rolling fast on a chain reaction style track, continuous smooth shot”
- “The video shows items of the alphabet. An unusual item starting with each letter is shown sitting on a table (like a Capybara for C, disco globe for D and Lava Lamp for L). All 26 letters must be represented by 26 items with matching lower thirds displaying the letter. Only one item and lower third at a time. Each lower third must look like a black marker written on a slip of paper in the bottom left. Rapid fire, roughly 9 frames per item at 24FPS. Last frame is a slip of paper THE END. The whole video is accompanied by calm smooth music”
The alphabet prompt is the best example of why this technique matters. The video runs 26 distinct items with consistent visual treatment (table, lower-third caption, marker handwriting) at a specific frame rate. The prompt does not describe what each item looks like; it relies on the model knowing what a capybara, disco globe, or lava lamp is. The 26-item structure is the part you have to specify; the visual realism is the part you can trust.
How do I control text rendering in the video?
Technique 2. Omni renders text well and lets you steer typography, spatial placement, animation, and effects like double exposure. Treat text as a first-class element of the prompt, not an afterthought.
Real prompts from the thread:
- Word-by-word sizzle reel: “word by word, one word on the screen at a time: did, you, know, that, this, model, can, do, pretty, good, text!? Each word appears with a different animated style, perfect pacing to a rhythm, sizzle reel”
- Motion-tracked intrusive thoughts overlay: “Overlay motion-tracked, minimalist text commentary onto the physical environment of the video. This text represents [the subject] deadpan, immediate inner monologue that’s observant, slightly absurd, and life-contemplating. Think intrusive thoughts. Clean, white, lowercase sans-serif text (like Helvetica or Inter). The text hovers in 3D space, connected to the subjects being commented on via ultra-thin, crisp, white leader lines”
The motion-tracked text examples are the most useful for content creators. The annotation pattern in the second image was generated from one reference image and one prompt, and it includes smooth transitions, accurate typography, and crisp leader lines. That is harder than it sounds; most AI video models still fail at small clean text and motion-tracked overlays.
How do I direct the camera like a cinematographer?
Technique 3. Use industry vocabulary. Omni was trained on enough film-school terminology to respond to specific cinematography terms, framing instructions, and camera-type cues. Vague directions (“zoom in slowly”) work less well than precise ones (“dolly zoom”).
Vocabulary the model responds to:
- Shots and angles: “one continuous shot”, “oner”, “static”, “locked off”, “fixed angle”
- Camera movements: “push in”, “punch in”, “pan left”, “dolly zoom”
- Camera styles: “natural smartphone zoom”, “vintage film camera”, “grainy webcam style”
The camera-style cues are the ones most beginners under-use. “Vintage film camera” gives Omni a whole package of look (warm tones, slight grain, slow shutter, period-appropriate lens behavior) in two words. Try those phrases before you reach for adjective-heavy descriptions.
How do I edit iteratively without losing my work?
Technique 4. Do not rewrite your prompt from scratch when you want to change one thing. Omni preserves the core structure across multiple targeted amends. Treat it the way you would treat any other edit: change the specific element, keep the rest.
Real iterative-edit prompts from the thread:
- “Transport the violin to a new environment”
- “Make the violin invisible”
- “Change the camera angle so it’s looking over the violinist’s shoulder”
The violinist screenshots show the technique working: the same character, the same dress, the same posture, but the environment shifts and the camera angle moves. Each amend changes one thing. This is how you should work with Omni for any sequence longer than a single shot.
How do I change action on the fly?
Technique 5. Modify a character’s pace, emotion, or movement mid-scene without breaking the continuity of the character model. Omni separates the character identity from the action being performed, so you can swap the action while keeping the person consistent.
Real action-modify prompts from the thread:
- “Make the character walk on their tiptoes”
- “Speed up the pacing”
- “Have them leap into the air”
This combined with technique 4 (iterative editing) is where Omni starts to feel like a film-set director’s chair. Set up the character. Modify the environment. Modify the camera. Modify the action. Each command is a small targeted change that preserves the rest.
What else has the community discovered Omni can do?
Three bonus techniques surfaced in the days after the Google AI thread that are worth knowing.
Style transfer across a whole shot
From Google Flow’s @FlowbyGoogle account: prompt “Transform the overall mood into a 3D anime style animation” applied to a real pigeon clip and produced a fully-animated version of the same scene with the same composition. Style transfer at the shot level, not just per-frame.
Cross-language audio translation
From @laszlogaal_: Omni can translate the audio of a video into another language without changing the speaker. It keeps the background music intact and adjusts the edit if needed. No translated text in the prompt; the model figures out the localized audio from the source video alone.
Environment swap with motion preserved
From @maxescu: “Add anyone to the Backrooms with Gemini Omni.” Prompt: “Keep the subject’s motion and timing exact. Place the subject in the Backrooms: yellow patterned wallpaper, damp musty carpet, low ceiling.” The character’s walk, posture, and timing transfer; the environment is fully rebuilt around them.
These three are the most actionable extensions of the five core techniques. For more on AI video model comparisons, see Sora vs Runway vs Kling and the AI Tools Directory.
Frequently asked questions
Is Gemini Omni free?
The Flash tier is free in the Gemini app and YouTube Shorts. Higher-quality and longer-form generation is rolling out behind paid plans (Gemini Advanced, Flow Pro). Pricing details will continue to update as Google ramps Omni rollout.
What is the difference between Omni and Veo?
Veo (now Veo 3.1) is Google’s standalone video model. Omni is the broader multimodal architecture that includes video as one capability. Practically, Veo is the deeper pure-video product; Omni is the every-modality entry point for casual users.
Can Omni clone real people?
Google’s policies restrict generating identifiable real people without consent. The Backrooms-style demos use anonymized or non-identifiable subjects. Treat character generation as creative-fiction territory, not impersonation.
How long can a Gemini Omni video be?
Omni Flash currently generates short clips (5-10 seconds typical). Longer-form work is built up through the iterative editing technique (technique 4) or by chaining clips in Flow.
Where can I see more community demos?
The @GoogleAI, @FlowbyGoogle, @icreatelife, @laszlogaal_, @patrickassale, and @maxescu accounts on X are the most active sources of Gemini Omni examples in real time. The Sources section below links to the original thread.
Get the daily Beginners in AI newsletter
One issue a day. Practical guides on AI video generation, prompting techniques, and the tools that actually work. Built for non-engineers.
Get Smarter About AI Every Morning
Free daily newsletter. Built for people who want to use AI well, not chase every model.
Free forever. Unsubscribe anytime.
Sources
- @GoogleAI on X: original Gemini Omni prompting guide thread (May 26, 2026)
- @FlowbyGoogle: style transfer and editing demos
- Gemini app (Omni Flash live here)
- Flow by Google (filmmaking surface)
- Google AI Blog (broader Gemini coverage)