Gemini Omni 1.2 Flash and the cost of revising product videos

Public developments and original analysis · Announcement date: 2026-11-05 · Sources

By the third round of revisions, a product-video team may already like the visual style. The remaining request is to change the background, adjust a shot or create a different aspect ratio for another placement. Regenerating the clip can also alter packaging, colours or product details that have already been approved. The editor watches again, the product owner checks again, and the time spent grows alongside the generation bill.

Google released Gemini Omni 1.2 Flash on November 5, with generation and editing of videos up to 15 seconds, improved reference following and multi-turn conversational editing. Google Cloud has announced general availability; rollout schedules differ across application surfaces and should be checked for the intended entry point. For teams producing product clips regularly, the interesting question is how revisions behave: can the next edit preserve approved content while changing the requested detail?

Relating generation prices to a finished video

Google announced an output price of approximately $0.07 per second for 720p video, 30% below Omni 1.1. The developer pricing table bills by output tokens and gives a 15-second example of approximately $1.04, with input charged separately. That prices one output. The cost of a finished piece also depends on attempts, edits, material preparation, post-production and review.

Suppose a team needs two placement variants for one product and generates three 15-second candidates for each. Using the pricing table's example, those six outputs alone total about $6.24. This illustration excludes input, later edits and other production costs. It also makes no assumption that three attempts will yield a usable result. Naming the number of candidates gives a budget discussion something concrete to work with; multiplying the single-output price by the number of final videos omits discarded attempts.

Google describes drafting at lower resolutions and increasing resolution after selecting a candidate. Direction, pacing and composition are sensible early decisions. Small packaging text, edges and material detail may need another inspection at delivery size. Each draft needs a purpose, or inexpensive candidates can create an expanding selection task.

Revision requests deserve separate consideration too. Changing a background, changing aspect ratio and correcting the product's shape involve different judgments. The first two can build on an approved reference version. Persistent product distortion calls for checking whether the supplied material explains the product sufficiently, or choosing filming or motion applied to still photographs. Repeating similar prompts may add little at that point.

Defining what a product edit may change

Consider a hypothetical social video for a shoe. The brand wants to retain the shape, colourway and sole pattern while trying two backgrounds for different audiences. Preparation should establish reference images of the same model and colourway, with the details that must remain. If one reference shows a sole from another production batch, the input already contains a conflict. Reviewers then have difficulty deciding which result is correct.

Google's editing guide recommends concise instructions for a specific change, together with a request to retain the remaining content. This helps limit the intended edit, although the actual output still needs comparison with the references. New lighting may shift the shoe's apparent colour; motion may distort laces or the sole. Both affect whether the clip is suitable for product presentation.

Creative setting and product claims also need different evidence. Showing a shoe in an imaginary landscape can be a clearly identified creative treatment. Images suggesting waterproofing, durability or performance in particular conditions need support from product information. Google acknowledges that fast action and complex physics remain areas for improvement. Performance demonstrations should favour material that represents real conditions, so a generated scene does not become apparent proof of an unverified claim.

Mokyun's AI marketing-content digital employees start from approved product information and brand material to prepare scripts, shot plans, suggested durations and required assets, with review comments attached to content versions. These tasks fit before generation: which specification supports a statement, which photograph shows the current model, how the setting may change and which claims still need confirmation. Video production and model invocation are separately agreed work. This service scope provides no evidence that Mokyun has already validated an Omni 1.2 integration.

Clear inputs make the model's contribution easier to assess. With the same approved material, a team can observe how much accepted content survives a background edit, whether the product owner must repeat a frame-by-frame inspection and how many candidates reach delivery. Output prices, revision effort and actual use together give a more useful account of production cost.

Choosing suitable clips for generation

Teams that regularly need different settings and have a consistent reference library have reason to try the update. Longer generation and continuing edits may reduce joins between clips and help refine an established creative direction. A simple subtitle update, or a requirement to reproduce packaging faithfully, may be served more directly by existing editing templates, filming or still-image animation.

Our preference is to test clips with a specific revision need and product details that can be checked. A successful edit preserves the accepted parts, allowing review to concentrate on the intended change. If each edit introduces another product discrepancy, lower generation prices offer insufficient grounds to expand usage. Keeping a more dependable production method may save time and better serve the requirement to show the product accurately.

• Mokyun marketing content and video-script preparation

Mokyun Editorial · neo@mokyun.com