Gemini video model gets 30% price cut: 15-second clips and conversational editing arrive together

Gemini Video Model Price Cut by 30%: 15-Second Clips and Conversational Editing Arrive Together

Gemini Omni 1.2 Flash has cut the price of 720p video to $0.07 per second. It can natively generate videos up to 15 seconds long and supports multi-turn conversational editing, reference-material consistency, and later high-definition output, so generation and modification can happen in the same workflow.

• Generating a 15-second 720p video costs $1.05

• The new model natively generates videos up to 15 seconds long and supports specifying length by the second

• Draft at 360p or 720p first, then upgrade to 1080p or 4K

• Multi-turn conversation can swap a product's background and change lighting and aspect ratio

This article is based on Google's official blog; pricing, duration, resolution, availability, and consistency results are all vendor-disclosed, and no third-party retesting was seen.

When operations staff face a batch of 15-second product short videos to be generated, with asset images, selling-point scripts, and a rework budget spread out on the desk at the same time, the first thing they calculate is not model parameters but how much each finished clip will cost. After Google released Gemini Omni 1.2 Flash, 720p video dropped to $0.07 per second, making a 15-second finished clip $1.05; along with the price cut, maximum generation length, reference-material consistency, and conversational editing were also upgraded.

How much does a 15-second video cost after the price cut?

The answer is a little over one dollar. Under the official new pricing, generating a 15-second, 720p video costs $1.05.

This price adjustment uses Gemini Omni 1.1 as the reference, with video prices down 30%. Google's new unit price is $0.07 per second for 720p video, billed by the second, so the longer the duration, the higher the cost.

For operators and creators producing short video ads, product demos, or virtual try-on content in bulk, $1.05 is the baseline generation cost for a 15-second 720p finished clip. Subsequent modification costs are not included, and the final delivery cost after upgrading to a higher resolution is calculated separately.

$1.05 is only the generation baseline: rework and HD upgrades are billed separately, and the real cost of bulk delivery depends on these two items.

What can the new model behind this price cut do?

Gemini Omni 1.2 Flash has been launched, and the most direct change is this: it can natively generate and edit videos up to 15 seconds long, and the duration can be specified in 1-second increments.

“Native generation” means the model produces the video directly, without first using other tools to generate assets and then splicing them into a finished clip. Gemini Omni 1.2 Flash puts both generation and editing into the same model, so the duration a user requests is no longer just an approximate range but can be specified down to the second.

Selectable durations run from 3 seconds to 15 seconds, in 1-second increments. A single piece of content no longer has to be limited to shorter clip lengths; creators can first decide the finished duration they need, then have the model generate a video of the corresponding length.

The previous generation, Gemini Omni 1.1, had a maximum duration of 10 seconds, and the new model pushes the ceiling to 15 seconds, an increase of 5 seconds. Official use cases include telling a complete story, making product explanations, and generating a complete social media ad clip in one go.

So what this round of updates brings is not video output that can only be watched and not customized, but a new model that can control length by the second while handling both generation and editing tasks. As for exactly how much this price cut reduces production costs, that still has to be considered together with the final generation cost of each video.

Draft in low resolution first, then push costs down to the final cut

To keep pushing generation costs lower, the key is not to directly produce a high-resolution finished clip every time, but to separate trial and error, correction, and finalization.

Google DeepMind's official workflow starts with a low-resolution draft: first generate an initial version at 360p or 720p, confirm the content and direction, and then make follow-up edits to the problems found. 360p and 720p are not finished-clip specifications but stage options for rapid drafting; the official materials have not published separate low-resolution billing rules, so it cannot be inferred from them that these are priced at a fixed ratio.

If the visuals look wrong at the draft stage, there is no need to regenerate the entire video from scratch. Creators can change only the erroneous parts and keep the content they are satisfied with. The official materials do not give a price list for editing and upgrading, so how much is saved depends on the actual number of reworks.

Multi-turn editing can cover the full 15-second clip, and creators can continue working on the details of the same asset without cutting it into shorter segments and splicing repeatedly. 15 seconds is the longest video clip natively supported by the model, and also the full length that multi-turn modification can cover.

Once the composition, subject, and modification results are all confirmed, upgrade the satisfactory work to 1080p or 4K. Computing resources are thus concentrated on the final version, and the proportion saved depends on how many times rework was done earlier.

How many versions can one generation produce?

Gemini Omni 1.2 Flash modifies videos using multi-turn conversational editing. Creators can keep adjusting along the current clip without having to regenerate from scratch every time a detail changes.

How many delivery versions can one piece of footage be edited into? Try it with a product ad: first swap the product, then move the environment from indoors to outdoors, then change the daytime lighting to nighttime, and finally switch the aspect ratio once—vertical for the phone feed, horizontal for a player or web page. All four steps are done on the same piece of footage, without starting a separate generation task for each change.

Multi-turn conversation is what handles the version branching. Keep the same product display and change only the background and lighting, and you can produce one version for each of different audiences; switch the aspect ratio again, and the same content can also enter different placements. One generation, extended into multiple market versions.

Google's official description confirms that Omni 1.2 Flash supports editing clips through multi-turn conversation and can directly correct details in subsequent adjustments. The vendor has not disclosed a uniform success rate or average time for multi-turn modification, nor has there been third-party retesting. What can be confirmed at present is the editing path and the adjustable scope; it is not yet possible to judge the stability of each modification.

How do you keep characters and brand details consistent across a whole video?

In a 15-second clip, does the character still look like the same person after a shot change? This is the hardest question for generative models to answer. Reference images, motion clips, and audio prompts are given to the model together, each constraining something different, and whether they can hold across the whole video is the key.

Look at reference images first. Creators can provide character images, product images, or style images at the same time: the character image defines the appearance, the product image constrains the look, and the style image sets the visual direction of the whole piece. Only with all three given together does the model have a basis for judging who should appear, what the product looks like, and what the shot should feel like.

What about motion? Text descriptions often cannot express it clearly. The model can receive motion clips up to 3 seconds long and reference the character movement or performance rhythm in them. The clip only provides a motion basis and does not go directly into the finished video—the finished video is still generated by the model, and the core performance in the clip can carry over into later frames.

Sound likewise participates at the generation stage. Audio prompts are given together with images and motion clips, and the model arranges sound cues accordingly, without waiting until after the finished video to find assets separately. With several reference streams constraining at the same time, characters, products, motion, and sound will not drift apart from one another.

Whether the references are actually followed ultimately depends on whether cross-shot details hold up. The officially stated scope includes faces, clothing, packaging text, and brand colors: after the same character transitions, face and clothing must connect; when a product cuts from front to side, the text on the packaging and the brand colors must stay consistent. These parts directly affect product demos, virtual try-ons, and series ads—if packaging text and brand colors do not hold, series ads cannot be used.

Source: Google Blog “Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips”; pricing, duration, resolution, availability, and consistency results are all officially disclosed by Google, and no third-party retesting was seen