Gemini Omni 1.2 Flash cuts video price by 30% and extends clips

By James Calloway

Google said it cut Gemini Omni 1.2 Flash pricing by 30%, putting 720p video at $0.07 per second. Google says the model is now generally available across the Gemini API and Gemini for business.

Google says the release raises native clip length from 10 seconds in Gemini Omni 1.1 to 15 seconds. It also adds multi-turn conversational editing (successive prompts that revise one clip) and stronger reference following (retaining supplied visual and audio cues).

These changes could reduce rework in advertising and product-video pipelines while producing longer assets in one request. Google supports them with demonstrations and unnamed adoption claims; it provides no independent benchmark scores or error rates for its consistency claims.

What changed in Gemini Omni 1.2 Flash?

Google says developers can draft at 360p or 720p, fix one detail through a follow-up edit, then upscale (increase the resolution of) selected clips to 1080p or 4K. Google presents the workflow as a way to avoid regenerating an entire clip after a minor correction.

Google says users can request a duration in one-second increments from 3 to 15 seconds. The company identifies complete stories, product explainers and full social advertising spots as work that can fit inside one generation.

Google's announcement gives no separate price for 360p, 1080p or 4K. The practical takeaway is that the headline rate applies to 720p, while the cost of the full resolution workflow remains unstated.

How closely does it follow references?

Google says generation can combine character, product or style images with short motion clips of up to three seconds and audio cues. It claims the model retains faces, garments, packaging typography and brand colors across camera angles, supporting virtual try-ons, catalog spins and branded video ads.

In one example, Google supplied a model photo and front-and-back product shots of three outfits. The model produced a 15-second virtual try-on across 360-degree turns and match cuts (transitions built around visual continuity). Google says it retained the model's identity and each garment's cut, drape and back details.

Google also says the model shows less visual drift or morphing across motion and camera transitions. It makes the same claim when interpolating motion between a first and last keyframe (two specified frames) or extending an existing clip into a longer sequence.

Google says products, characters, props and environments remain consistent from shot to shot. It also says characters, text and camera moves stay consistent across longer shots, multiple turns and transitions, with no cuts.

Google's poster demonstration offers its clearest stated comparison with Gemini Omni 1.1 Flash. Given the same static studio-lighting exhibition poster and five-layer animation prompt, it says the older model matched the reference less closely.

In that sequence, Gemini Omni 1.2 Flash builds the image on a dusty-pink cyclorama (curved studio backdrop) above a rising hexagonal plinth. It assembles ribbed glass, warm filament, faceted crystal and coiled-cord letterforms for HALO, GLOW and LUMEN, then holds the materials, shadows and typography through the final frame.

That demonstration is not a benchmark. Google disclosed no scores, sample sizes, test conditions or independent evaluations for reference consistency, identity retention or typography. The takeaway is that reference control is the product's main pitch, not yet a measured result.

What can conversational editing change?

Google says a single 15-second clip can be revised turn by turn without starting over. Google says prompts can swap a product or background, change lighting or weather, or switch the aspect ratio.

Google positions the process as a way to create advertising variants for different audiences and placements without re-briefing or regenerating the original clip. Another demonstration moves the same scene through five backgrounds: a striped pavilion tent, an oak forest, a stable with a white horse, a market fair and a stone castle.

Google says the model preserves both characters' performances, armor reflections and camera framing throughout. That sequence is a controlled example rather than a measured test, and Google does not quantify how often a revision changes an element the user asked to preserve. The takeaway is that conversational editing could reduce repeated generation, but preservation accuracy remains unverified.

How does it handle speech and camera movement?

Google claims Gemini Omni 1.2 Flash produces more fluid dialogue, clearer lip-sync (matching mouth movement to speech) and distinct multi-speaker conditioning (assigning separate voices to characters). It says those improvements apply to scenes in Flow, explainers in Gemini for business, and avatar, music and narration projects in Google Vids.

In one demonstration, the model receives one conservatory scene image plus four references covering character appearance, mouth shape and one-second voices. Google says it routes each sampled voice to the correct character while maintaining expressive lip-sync and natural comedic timing across a 15-second animated scene.

Google says the model can direct continuous first-person-view (FPV) drone flights, sweeping flythroughs, dramatic zoom-outs and orbits with significantly fewer unwanted jump cuts. The company also reports steadier scale and a more polished, cinematic result.

Guided by two architectural reference photos, the model directs one continuous FPV flight across a piazza, through bronze doors and up through the Pantheon's coffered dome to the oculus. Google says the shot contains “zero jump cuts.”

The announcement supplies no numerical measures for dialogue quality, lip-sync accuracy, camera continuity or physical realism. Independent testing would need to establish whether those improvements hold outside Google's chosen examples. The takeaway is that Google has demonstrated smoother-looking examples without establishing how often they fail.

What does Google say about enterprise use?

Google says businesses, startups and creatives already use Gemini Omni 1.2 Flash in their workflows. It also says Gemini Omni is generally available for the first time, backed by Google Cloud's enterprise-grade infrastructure and security.

Google says Provisioned Throughput, a service for reserving model capacity, is available. It describes every output as carrying an “imperceptible” SynthID watermark (an embedded content marker).

Google did not name the businesses, startups or creatives it cited. It also disclosed no throughput levels, service limits, watermark-detection performance or comparative operating data. The takeaway is that enterprise availability is established more clearly than enterprise performance.

Where is it available, and what remains unfinished?\n

Google says developers can start in Google AI Studio, use interactive starter apps and consult the developer documentation and pricing. Businesses can test the model in the Google Cloud console and build enterprise solutions through the API.

Google says users can direct scenes and make multi-turn edits in Flow. It says Google Vids will generate 15-second clips at no additional cost for people with a Google account or Google Workspace subscription.

Google says the Gemini app is rolling out 15-second video creation to Google AI Ultra subscribers. It also says Google Ads users can scale video campaigns with the model.

The access paths do not carry equal status. The API and Gemini for business are generally available, while the Gemini app rollout remains limited to Google AI Ultra subscribers, with no timetable for wider access. Developer access is therefore clearer than mass-market availability.

Google describes the model as a practical fit for businesses, developers and creators producing video at scale. It says development continues on fast sports action, rapid script changes and complex physics.

Google has not quantified the model's limits in those areas. The next test is whether independent users can preserve references and revise 15-second clips under real production constraints.