Google announced Gemini Omni 1.2 Flash on November 5, offering videos up to 15 seconds long and editing through successive conversational instructions. The company describes general availability through the Gemini API and an effective price of approximately $0.07 per second of 720p output. This is a new video-generation model version, rather than a model for analyzing existing footage. For professional users, the change concerns both clip length and the workflow used to refine a result.
Duration, audio and workflow
The technical video guide directs developers to gemini-omni-1.2-flash. It describes generation of 3-to-15-second clips with native audio, followed by conversational editing and extension through the Interactions API. Understanding input video is a separate task with separate documentation. The distinction matters when selecting a model and deciding what a successful result should demonstrate.
Google claims improved visual-reference adherence and more consistent objects and characters. These are provider claims rather than findings from our own comparative testing. Accepting a reference does not establish that every detail will be reproduced accurately in every frame. The properties that matter to a particular project still need to be assessed in its output.
For a team creating a product clip, conversational editing would be valuable if it changed one requested element while preserving the others. Whether that works sufficiently well for a particular catalogue requires checking with the team’s own material. The intended workflow reduces the need to restart, but it does not remove the need to inspect the result.
Interpreting the per-second price
The official pricing page places the model on the Gemini API paid tier. Output is billed by tokens; the documented effective rate for 720p is approximately $0.07 per second. Inputs have a separate price. That figure is therefore not a guaranteed final invoice for a clip with arbitrary references, revisions and resolution choices.
This matters when comparing providers. An accepted clip may require several attempts, so the budget should include variants and revisions rather than only the duration of the final file. A lower unit rate does not automatically establish a lower cost per approved advertising asset. The relevant denominator is the result a team can actually publish.
Our editorial recommendation is to compare tools with one small, fixed task. Preserve the input images, prompt, requested duration, resolution and permitted edits. Record time and cost across every attempt. That measures the team’s own production workflow without turning a vendor’s consistency claim into a general guarantee. It also makes a later comparison reproducible when the model or settings change.
What to inspect frame by frame
For product video, define in advance which properties must remain unchanged: shape, color, proportions and placement of distinctive elements. Separately assess movement, transitions and agreement between audio and visuals. An impressive opening frame is insufficient when the asset must remain correct throughout the clip.
For example, an edit intended only to change lighting might also alter the product’s appearance. This is our illustrative acceptance-test scenario, not a reported Omni defect. The comparison should show whether the requested change occurred and whether the properties intended to be preserved really stayed the same. A visually pleasing revision can still fail that narrower requirement.
Use material with appropriate rights and distinguish a synthetic demonstration from recorded product behavior. In explanatory videos, visual persuasiveness can exceed factual accuracy. Human review therefore needs to cover the meaning of what is shown as well as its aesthetic quality, particularly when the clip illustrates a function a customer may rely on.
What follows from the announcement
The checked public documents establish the model version, maximum duration, described API workflow and billing approach. We have not generated comparative videos or independently confirmed the claimed quality improvements. Availability of individual functions should be checked for the specific product and account rather than assuming every demonstration implies identical access everywhere.
The practical value of Omni 1.2 Flash will be measured by how much work it takes to reach an acceptable finished clip. Longer output offers more room for a story while also creating more frames to inspect. The balance between control, repeated attempts and a verifiable result determines the release’s usefulness to a production team.
Editorial verification: 2026-11-06 19:32 (Europe/Sofia).
