Gemini Omni Flash Video: Google's 4K Model, Priced Low

Short version: Google published Gemini Omni Flash video on 27 August 2026. Four resolution tiers from 360p to 4K, clips of 3 to 10 seconds, and a 360p tier at $0.30 for five seconds that makes it the cheapest way to test a shot on this shelf. Only two aspect ratios, which is its one real limitation.

This model is a day old. There is not yet a body of real-world output to judge it on, so what follows is the specification, the price, and where it sits, without pretending to more.

Disclosure: Leaxor is mine and now runs it.

What Google shipped

Gemini Omni Flash sits in the Gemini Flash line, which is Google's speed-and-cost tier rather than its flagship. The video endpoints arrived on fal on 27 August alongside image-to-video, reference-to-video and an edit mode.

Google announced the Omni Flash line in its own developer post. The specification is narrow and deliberate:

• Resolutions: 360p, 720p, 1080p, 4K. Default 720p.
• Duration: 3 to 10 seconds. Default 8.
• Aspect ratios: 16:9 and 9:16 only.

Two aspect ratios is unusually restrictive. Most models here offer five or six. If your work is landscape or vertical you will not notice; if you needed square, this model is simply not for you.

The price spread is the interesting part

Four tiers, and a tenfold gap between the cheapest and the dearest. That is a wider spread than any other model on the shelf.

$0.30 for five seconds at 360p is the cheapest render available here. It looks like what it costs, and that is fine, because nobody publishes a draft. Use it to answer "does this framing work" across a dozen attempts for $3.60, then spend once at the top.

The settings, and the two that constrain you

This model has a short parameter list, which is part of the point. Three things to know.

Duration is 3 to 10 seconds, default 8

That upper bound is the real constraint. Ten seconds is one shot, not a sequence. If your idea needs a beat, a turn and a reveal, you will be stitching multiple generations and dealing with the continuity yourself.

The default of 8 is also worth overriding. Most short-form cuts want 3 to 5 seconds, and generating 8 to use 4 is paying double for the trim.

Two aspect ratios, and that is the whole list

16:9 and 9:16. No square, no 4:3, no 21:9.

For YouTube and Shorts those are exactly the two that matter, so most people will never notice. If you post to a square format, or you were hoping to crop from something wider, this model simply is not for you and no setting will fix it.

Resolution is a ninefold price range

360p, 720p, 1080p and 4K. The gap between the bottom and top of that ladder is larger than on any other model here, which makes picking correctly worth more than usual.

The practical rule: 360p until you know the shot works, then jump straight to whatever the destination actually displays. Which for vertical social is 1080p at most, and often less.

What it is genuinely for

Draft passes. At $0.30 this is the cheapest way to find out whether an idea survives contact with a video model. Generate five variations of a prompt, look at them side by side, throw four away.

Single hero shots at 4K. Very few models offer 4K at all. If you are making something that will be seen on a large screen, this is one of the only routes, and $3.00 for five seconds is not a lot for that.

Reference-driven work. The reference-to-video and edit endpoints are the more interesting half of this release and go almost unmentioned elsewhere. They let you steer from existing media rather than from a sentence, which removes most of the randomness.

What it will not do is carry a sequence. Ten seconds and two aspect ratios is a shot generator, and a good one.

Where it fits

Two genuine strengths. The 360p tier is the cheapest exploration on the shelf. The 4K tier exists at all, which is rare.

Two genuine limits. Ten seconds is short next to Wan 3.0's thirty. And two aspect ratios is two.

Against MiniMax H3 Max the shapes differ usefully: H3 Max is cheap in the middle and caps at 768p, Gemini is cheaper at the bottom and reaches far higher at the top.

Four modes, and only one of them gets written about

Everything published about this model so far covers text-to-video. It shipped with four endpoints, and the other three are the more interesting half.

Image-to-video

Start from a still and describe how it should move. This is the mode serious users lean on, because you keep control of composition, framing and style, and hand over only the motion. Text-to-video asks a model to invent a shot; image-to-video asks it to animate one you already approved.

The practical workflow that falls out of that: generate the frame on a cheap image model, judge it, then animate the winner. You are choosing the picture for a fraction of a cent instead of rolling for it at video prices.

Reference-to-video

Takes reference media to steer the result rather than a start frame to animate. It is a third input shape and most tools do not model it at all.

It matters for consistency. If you have an established look, feeding it as reference is more reliable than describing it in prose and hoping the adjectives land the same way twice.

Edit

Change an existing clip rather than regenerating it. The distinction is the same one that makes region-precise image editing valuable: regenerating means everything you liked is also up for renegotiation.

Why the extra modes decide the model

On text-to-video alone this is a competent cheap model among several. With reference and edit endpoints it covers a workflow rather than a step, and that is a different kind of usefulness than a leaderboard position measures.

Worth trying?

At $0.30 a draft, the question answers itself for anyone already generating video. Whether the 4K tier is worth $3.00 a clip depends entirely on whether anything downstream will ever display 4K, and for short-form social the answer is usually no.

Gemini Omni Flash and its image-to-video and reference modes are live in Leaxor's clip generator, priced per render, credits at $0.10 and no subscription required. The full August wave is in the roundup, and every model's rate is in the cost breakdown.

Reading this with an AI assistant? Ask it to summarize this in ChatGPT, Claude, or Perplexity.