Loading page…
The update lets users revise clips through follow-up prompts. Google promises better visual consistency, but acknowledges remaining weaknesses in fast action and complex physics.
Listen to this story
Google's November 5 release makes Omni 1.2 Flash more practical for iterative video production: users can revise clips across their full duration and upscale selected drafts instead of regenerating from scratch. The workflow targets ad variants and product videos, where preserving people, packaging, framing and voice could reduce rework. Google's demos support these quality claims, but they are not independent validation; the company still flags fast sports, rapid script changes and complex physics as weak spots.
720p generation costs $0.07 per second, 30% less than Omni 1.1; native clips run from three to 15 seconds in one-second increments.
Users can draft at 360p or 720p, refine clips, then upscale selected results to 1080p or 4K.
Google claims stronger reference-following and continuity, but its examples include small changes such as a soup ladle turning into chopsticks.
Google is promising cheaper AI video that is easier to revise without losing visual details. Released November 5, Gemini Omni 1.2 Flash brings 15-second clips and a 30% price cut. Those changes are concrete; its stronger consistency is a company claim illustrated through demos, while Google acknowledges that fast sports action and complex physics still need work.
The model is generally available through the Gemini API and Gemini for business. Google lists 720p generation at $0.07 per second, down 30% from Omni 1.1. It also raises the native clip limit from 10 seconds to 15 seconds, with durations selectable in one-second steps from three to 15 seconds.
Conversational editing works across the full 15-second clip. Users can ask for a different background, swap a product, change lighting or weather, or switch the aspect ratio through successive prompts. Google positions this as a way to make ad variants without starting again or repeating the original instructions.
One company demonstration takes a generated scene through several background changes, moving from a striped pavilion to a forest, stable, market and castle. Google says the characters’ performances, armor reflections and camera framing remain intact. That is the central editing promise: change the requested element while preserving the rest of the shot.
The release also supports a draft-and-finish workflow: generate at 360p or 720p, refine the clip, then upscale selected results to 1080p or 4K. Google argues that follow-up edits, rather than complete regenerations, can help keep production costs down alongside the lower listed generation price.
Users can guide generation with a combination of character, product and style images, motion references up to three seconds long, and audio cues. Google says the model follows those references more closely, preserving faces, clothing, packaging text and brand colors across camera angles. These are company-supplied quality claims, rather than independently established results.
In a virtual try-on example, Google combines one model photograph with front-and-back product shots of three outfits. It says the resulting sequence preserves the person’s identity and each garment’s cut, drape and back details through turns and cuts between shots. The example targets a specific business use: turning existing product photography into video.
Google also claims more natural dialogue, clearer lip-sync and better control over which voice belongs to which character. For camera direction, it promises smoother continuous movements with fewer unwanted cuts and steadier scale. These improvements address speech and motion within a shot, not just how closely an opening frame matches a reference image.
The company’s qualifications are specific: fast sports action, rapid script changes and complex physics remain areas under development. Its first-to-last-frame comparison also notes small changes, including a soup ladle turning into chopsticks. Better consistency does not mean every object survives a generated transition unchanged.
Alongside developer and business access, the model is rolling out through Google’s video and advertising products. The launch announcement describes different entry points rather than identical access across every app:
For production customers, Google says Provisioned Throughput is available through its cloud infrastructure. Generated outputs also carry an imperceptible SynthID watermark, adding a marking mechanism to the videos delivered through the release.
Story updates
A Google account is enough to generate up to three 15-second AI videos a month in Google Vids at no cost, starting November 5, 2026. The video creation and editing tool now runs on Omni 1.2 Flash, which Google says improves how uploaded images carry through into moving scenes and how narration sounds.
The Vids update follows Google's launch of Gemini Omni 1.2 Flash earlier the same day. The underlying model raises native video generation from 10 seconds with Omni 1.1 to 15 seconds. It also supports durations in one-second steps, from three to 15 seconds, rather than requiring every clip to use the maximum length.
Google pitches the longer Vids clips for short product promotions, social posts and explainers. Its announcement emphasizes making a promotional sequence in one generation without a cut. That is a proposed use for the feature, not a guarantee that every generated clip will be ready to publish.
The free allowance applies to users with a Google account or a Google Workspace subscription. Google directs people to vids.new to try the editor. The offer is capped at three generations per month; it is not unlimited access to the new model through Vids.
In Vids, users can upload reference images to guide the resulting video. Google suggests pictures from trips, school projects or learning materials, and says characters and scenes will remain consistent from the first frame to the last. It also says improved speech narration should make social clips, concept explainers and teasers sound more natural.
The broader model announcement explains what Google means by better reference following. Omni 1.2 Flash can combine character, product and style images with motion clips of up to three seconds and audio cues. Google claims it better preserves faces, clothing, packaging text and brand colors as the camera angle changes.
One company demonstration combines a model photograph with front-and-back product shots of three outfits to produce a virtual try-on sequence. Google says the generated video preserves the person's identity and the garments' cut, drape and back details through turns and cuts. Those are company demonstrations and claims, not independent verification of product accuracy.
Beyond the Vids-specific announcement, the new model supports conversational editing: users refine an existing clip through follow-up prompts instead of generating the whole scene again. Google says this works across a full 15-second clip. Its examples include several kinds of revision:
Swap a product or replace the background.
Change the lighting or weather.
Switch the aspect ratio—the shape of the video frame.
In one demonstration, successive prompts move the same scene between a pavilion, forest, stable, market and castle. Google says the characters' performances, armor reflections and camera framing stay intact. The editing promise is selective change: alter the requested element while keeping the rest of the shot recognizable.
That control still has limits. Google acknowledges that fast sports action, rapid script changes and complex physics need more work. Its own first-to-last-frame example also notes small changes, including a soup ladle becoming chopsticks. Longer clips and steadier references do not amount to perfect continuity, even in the company's selected examples.
blog.googleGoogle’s Gemini Omni 1.2 Flash is now available in Runway. Create and edit videos up to 15 seconds, with more natural dialogue and stronger consistency across shots. Refine your scene without starting over, then upscale to 4K. Try it now in Runway.
Loading discussion...
Join the conversation
Explain which details you would forgive—and which would undermine trust.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.