Google's Gemini Omni 1.2 Flash Adds 15-Second AI Video

Google's Gemini Omni 1.2 Flash can generate clips up to 15 seconds long and edit them through successive instructions, with 720p output priced at $0.07 per second. That puts a full-length clip at $1.05 if every second is billed at the stated rate, before any other charges.

Google has introduced the model as a faster, more cost-effective option for making longer clips. The reported price is a 30% reduction, while the new duration is 50% longer than before. Longer clips and cheaper seconds are straightforward gains on paper. Whether creators get more usable footage per dollar depends on how the model handles revisions.

Editing Without Starting Over

Conversational, multi-turn editing means a person can give a sequence of requests: generate a clip, change one element, then refine it again. Think of asking a friend to move a chair in a room they have already arranged. You expect the chair to move; you do not expect the walls to become purple or the dog to disappear.

For AI video, that expectation is a substantial technical challenge. A request such as “make the jacket red” may need to preserve the subject's face, the camera angle, the lighting and the timing of their movement. On the next turn, “slow down the final three seconds” should leave that red jacket in place. If those details drift, the user spends the time saved by conversational editing repairing the conversation's earlier results.

This is where the 15-second limit becomes more than a number in a product announcement. A longer clip gives an action room to develop: someone can enter a frame, pick up an object and react to what happens next. It also gives the model more opportunities to lose track of that object, change its appearance between frames or misunderstand which moment the user wants edited. Duration increases the space for a story and the space for errors.

The reported specifications include no shared quality benchmark for how reliably details survive successive edits. A useful comparison would start with identical prompts, require the same changes over several turns and score both what changed and what stayed put. A beautiful first draft would earn little credit if the requested final adjustment rewrote the rest of the scene.

The Price of a Usable Clip

At $0.07 per second, a 15-second 720p generation costs $1.05 under a simple per-second calculation. Three full-length generations would cost $3.15 at that rate; ten would cost $10.50. Those figures illustrate the stakes of retrying a clip, not a claim about how Google bills edits or failed attempts. A project's total may also depend on charges beyond the stated generation rate.

The 30% price cut has a similarly important qualifier: its value depends on what the previous rate covered. If the comparison uses the same resolution and billing unit, $0.07 represents a drop from $0.10 per second. That arithmetic helps put the reduction in perspective without assuming that every user's workflow gets 30% cheaper.

Picture two creators making the same 15-second shot. One gets an acceptable result on the second attempt. The other needs eight attempts because a character's clothing changes mid-clip. At the stated rate, their full-length generations would cost $2.10 and $8.40, respectively, even though both end with one usable shot. Generation cost is the menu price; retries are the bill that arrives when the kitchen keeps remaking the order.

This makes multi-turn editing a potentially valuable feature for developers and everyday creators alike. If a targeted instruction fixes a shot while preserving everything else, people can refine their work instead of repeatedly generating whole clips from scratch. If an edit introduces new problems, the conversational interface becomes a very polite way to ask for another retry. Billing rules for edits would shape that calculation too: a small change is more attractive when its cost reflects the work requested.

Google's wider Omni Flash API access gives developers a reason to care about these details beyond the creation screen. A tool built around repeated edits needs predictable behavior and predictable costs. Otherwise, its developer has to decide whether to absorb retries, pass them on to users or limit how often people can experiment.

Who Benefits from Fifteen Seconds?

A 15-second ceiling suits short demonstrations, transitions and individual shots. It does not automatically produce a longer coherent scene. Someone building a minute-long sequence would need to assemble several clips and manage continuity across them: the same subject, setting and action from one segment to the next. The new duration could reduce the number of segments required, while leaving that assembly work in place.

Different users will also judge “good” differently. A social post might tolerate a strange hand movement if the image otherwise delivers the joke. A product demonstration cannot casually change the shape of the product halfway through. A creator trying to match an existing visual style may care more about repeatable camera framing than about generating a surprising first draft. One headline rate cannot capture all three workloads.

For people evaluating video tools, the most revealing test may be deliberately boring: generate a person walking through a room, then change the wall color, then adjust the lighting, then shorten the ending. Count how many requested changes work. Count how many unrelated details move along with them. Finally, count the seconds generated to reach the usable version. That measures the workflow Google is asking people to try, rather than rewarding a single lucky output.

Google's update makes one promise easy to price: up to 15 seconds at a stated 720p rate. The promise of conversational editing is harder to put on a price card. Its payoff comes when “change the ending” changes the ending and leaves the rest of the clip alone.