Gemini Omni 1.2 Flash: Describe What You Want, Skip the Timeline
Google dropped Gemini Omni 1.2 Flash on November 5, and video creators are already rethinking how they edit. Instead of dragging clips around a timeline, you can now describe a change—"make it rain," "change the bottle color to silver," "tighten the pacing here"—and the model remixes the shot in one generation.
It's a small shift that removes a major friction point. Traditional AI video meant committing to a single prompt. If you wanted variants or tweaks, you generated a new clip from scratch. Omni 1.2 Flash lets you iterate on what you already have, preserving the elements you want to keep while transforming the rest.
What people are making
The early adopters are leaning into speed. Creators are spinning up quick product videos—shoes recolored across all three shots of an ad, a perfume bottle with fresh lighting and mist without reshooting, a night-market establishing shot with precise timing. One example showed a 15-second running-shoe ad with three distinct shots generated from a single prompt; when the color needed to shift to electric blue, the model kept the composition and character motion intact and just changed the hue across all three shots.
The editing capability is built for this. You can feed it a reference image of a vase and have it perform a monologue while holding that object, or upload a full clip and ask for a time-of-day swap. The model understands spatial relationships—it knows what to change and what to preserve.
Small teams testing it are reporting faster turnarounds for social content and internal marketing. No more waiting days for revision rounds. Text to video in 15 seconds, or edit existing footage in a few minutes.
What's actually good (and what isn't)
The real win is the free tier. Google Vids now lets any account holder generate three 15-second videos per month for free, using Omni 1.2 Flash. That's a genuine entry point for people who've never touched AI video.
The 30% price cut makes the paid tiers affordable too—720p video now costs $0.07 per second, so a 15-second clip runs about $1. That's lean for professional workflows.
Duration and consistency matter here. The model can hold a single unbroken shot for the full 15 seconds—no jump cuts forced by model amnesia. Testers report smooth 15-second drone shots with zero cuts, a historically hard problem in AI video. Character and object consistency across shots is better than previous generations, though still imperfect on edge cases.
The limitation is length. 15 seconds is enough for ads and social clips, but not for anything longer-form. You're also bound to what the model decides to do with your reference images or character references—there's no keyframe-by-keyframe control. And while the editing capability is potent, it still works best on straightforward transformations (color, time of day, weather). Complex compositional changes don't always translate cleanly.
How to try it
For free experiments, go to Google Vids and log in with any Google account. You get three 15-second generations per month. The interface is built for talking to the model—type what you want, describe references, ask for variants.
For API access and higher volume, Gemini Omni 1.2 Flash is available through the Gemini API and Vertex AI. Developers can script generation runs and layer them into existing workflows. Adobe is integrating the model into Firefly for in-app video editing, so if you're already in Adobe's ecosystem, that integration should roll out soon.
The conversation-based editing is the real test. Prompt it to change one shot and leave others alone. Ask for color variants. See if the model tracks your intent across multiple requests. That's where the friction relief actually lands.