The Ultimate Google Omni Prompting Guide (Full Tutorial)
This tutorial covers five key strengths of Google's Gemini Omni video model that most users overlook. The presenter demonstrates real video editing, camera movement manipulation, multilingual translations, real-world knowledge generation, and text rendering capabilities. Specific prompts and workflows using Google Flow are shared throughout.
Summary
The presenter argues that most users only tap about 10% of Gemini Omni's potential by using it solely for AI avatar content. The tutorial focuses on five underutilized strengths, walking through each with specific prompt examples and demonstrations.
The first strength covered is real video editing. The presenter explains that Omni can be accessed either through the Gemini app or Google Flow, with Flow being preferred for its iterative workflow. Using a beach video as a running example, they demonstrate how a single prompt can add a crowd to an empty beach, and how subsequent prompts can be chained — editing the previously generated video rather than the original — to stack changes like adding sunny weather.
The second strength is camera movement. The presenter shows how Omni can transform a ground-level beach video into a convincing drone shot with a simple zoom-out prompt. They also demonstrate a viral technique where arrows drawn onto a still image direct Omni's virtual camera path through a scene, producing smooth drone-like flythrough footage.
The third strength highlighted is multilingual translation, shown through birthday message videos generated in French, Spanish, Latin, Vulcan, and ASL using the presenter's avatar. French and Spanish outputs were verified as accurate via Google Translate.
The fourth strength is real-world understanding, where Omni generates informative explainer videos — complete with visuals, narration, and avatar overlays — from minimal prompts like 'explain how rockets work.' The presenter also demonstrates Omni's ability to replace a car's driving environment using Google Maps screenshots, convincingly replicating Manhattan and London street scenes while preserving interior car details.
The fifth strength is text rendering within videos, demonstrated by uploading a flower video and prompting Omni to add 3D-locked label overlays that identify different parts of the orchid as the camera moves. The presenter closes by acknowledging Omni's imperfections while emphasizing that understanding its strengths leads to impressive results.
Key Insights
- The presenter argues that iterating on generated videos — feeding Omni's output back as input for further edits — is the true strength of the tool, not one-shot generation, because it allows stacked, precise changes over multiple prompts.
- The presenter demonstrates that Omni can follow directional arrows drawn onto a still image to produce smooth, continuous drone-like camera movements through a scene, a technique the presenter notes has been going viral.
- The presenter claims Omni's multilingual translation capability is strong because it stems from the model's broader real-world understanding, not just language training, allowing it to generate contextually accurate video in languages like French and Spanish.
- The presenter shows that uploading a Google Maps screenshot alongside a driving POV video allows Omni to convincingly replace the driving environment with the mapped location — like Manhattan or London — while preserving interior car details like the dashboard, rear-view camera, and window stickers.
- The presenter notes that when Omni consistently fails to execute a desired transformation across multiple attempts — such as timing a water bottle-to-rubber-chicken swap — it is sometimes better to restart with a revised prompt rather than continuing to iterate on flawed generations.
Topics
Full transcript available for MurmurCast members
Sign Up to Access