Gemini Omni

Trademark and affiliation: Gemini is Google's product name. Gemini Omni is an independent third-party service that provides access to Google's models; it is not operated by, endorsed by or affiliated with Google. We found no affiliation disclaimer on the site's homepage. Verify who you are contracting with before entering payment details.

There is a category of AI product that deserves to be understood clearly before anything else: the reseller. A company that does not build a model but sells access to someone else's, wrapped in its own interface and its own brand.

That is a legitimate business and it can add real value — a better interface, useful defaults, features the original provider does not offer. It also means you are paying a margin above the underlying cost, and that your service depends entirely on a relationship you cannot see.

Gemini Omni is in that category. The generation capability is Google's; the interface, the credit system and the conveniences are this company's. Judging it fairly means separating those two things, because the model is strong and the question is what the wrapper is worth.

What Is Gemini Omni Selling?

• Access to Google's multimodal video generation through an independent interface — this is a reseller, not a Google product.
• Up to fifteen multimodal references per prompt, mixing text, images, video and audio, which is more than most interfaces allow.
• Native audio synchronised with visuals rather than added afterwards, plus character consistency from reference photos.
• In-chat editing without regeneration, so a change does not mean starting over and spending again.
• Output up to 4K in four to ten second clips; commercial use is restricted to the Pro plan.

Fifteen References Is The Real Convenience

Gemini Omni allows up to fifteen multimodal references per prompt, mixing text, images, video and audio.

This is the most substantive thing the interface adds, and it addresses a genuine constraint. Generating a shot that matches a specific look usually means supplying several inputs — the character, the setting, the style, a motion reference — and most interfaces cap you well below that.

The practical difference shows up in commercial work. A brand video needs the product, the palette, the typography treatment and a tonal reference simultaneously, and each reference you cannot supply is a quality you have to describe in words and hope for.

Character consistency from reference photos works on the same principle and solves the problem everyone hits in generative video: the same person looking like a different person between shots.

The caveat is that reference-conditioned generation is guidance rather than instruction. Supplying fifteen references does not mean all fifteen will be respected equally, and the model resolves conflicts between them in ways you cannot control. Expect strong influence, not obedience.

Reference-image handling is the capability worth comparing directly against going to Gemini yourself, since convenience around exactly this is what you are paying a reseller for.

A reference you cannot supply becomes a quality you have to describe in words and hope for. That is why the count matters.

In-Chat Editing Changes The Economics

Editing in the conversation without regenerating is the feature that matters most for anyone paying per generation.

The normal generative video loop is expensive in exactly the wrong way. You produce a clip that is nearly right, want one thing changed, and the only route is a new generation — which costs the same as the first and may lose what you liked about it.

Conversational editing avoids both problems. You keep the parts that worked and adjust the part that did not, and you are not gambling a full generation on a small change.

For a credit-based product this is the difference between a tool you use freely and one you use nervously. It is also worth testing specifically during a trial, because how much can actually be changed without regeneration varies enormously — some systems can adjust colour and pacing but not composition.

What Are You Paying The Reseller For?

This is the central question for any product in this category, and it deserves a direct answer rather than avoidance.

The honest case for a reseller is real. Google's own developer surfaces assume an engineer; a web interface with credits, a gallery and conversational editing does not. For someone who wants to make videos rather than write code, that gap is worth something, and simplifying access to a frontier capability is a genuine service.

The honest case against is equally real. You pay a margin above the underlying model cost, because the reseller has to make money on the difference. You get whatever capability the reseller has implemented rather than everything the model can do. And your access depends on a commercial relationship between two companies that you have no visibility into.

That last point is the one people underweight. If the arrangement changes — pricing, terms, availability — the effect reaches you with no warning, and there is no recourse because your contract is with the intermediary.

The practical guidance: use a reseller for convenience on discretionary work, and go direct for anything a business depends on. And before paying, spend two minutes finding out who the company actually is.

The question generalises. Whenever an interface sits between you and a model, ask what it adds beyond access — the answer for InVideo AI, for instance, is an entire editing workflow rather than a thinner wrapper.

• Establish who you are contracting with before entering payment details.
• Check whether the same capability is available directly, and at what price.
• Read the refund policy carefully, since credits are prepaid and disputes are common in this category.
• Confirm commercial rights explicitly, since here they sit only on the top plan.
• Do not build a business process on a reseller relationship you cannot see the terms of.

The Output Specifications Are Genuinely Published

To the site's credit, it publishes specifications that many competitors omit: up to 4K resolution, four to ten second durations, native stereo audio synchronised with the visuals, and fifteen multimodal references.

Clip length is the number to read carefully. Four to ten seconds is standard for current generative video and it constrains what you can make: a social clip, a product shot, an establishing sequence. Anything longer means generating several clips and editing them together, which brings back the consistency problem the reference features exist to solve.

Native audio sync is more valuable than it appears. Video generated silently and scored afterwards has a characteristic disconnection — footsteps that do not land, an impact with no matching sound. Generating audio with the visuals rather than over them is a meaningful quality difference.

4K output is at the top of what current systems produce, and worth treating with the usual caution: highest resolutions are typically slower, more expensive, and sometimes less reliable than mid-range ones.

Publishing these at all is better behaviour than most of this category manages, and it lets you judge fit before spending anything.

Published specifications also make honest comparison possible with platforms like Runway ML, where you can hold resolution and duration claims against each other rather than guessing.

Testing It Against Going Direct

Ten free credits on signup make this testable, and the comparison worth running is not against other resellers but against the underlying model itself.

The bar is step four. If the interface does not save you meaningful effort over direct access, you are paying a margin for nothing.

1. Read the terms and refund policy first, and establish who the operating company is. For a prepaid credit product this is not optional.
2. Use the free credits on a shot you actually need, supplying several references rather than a text prompt alone.
3. Test in-chat editing specifically: change one element and see how much can be adjusted without a full regeneration.
4. Generate the same shot through Google's own surface if you have access, and compare quality, effort and cost honestly.
5. Check the commercial terms before using anything in paid work, since commercial rights are stated as Pro plan only.
6. Only buy credits if the interface genuinely saved you time, since the model is the same either way.

How The Credit Plans Work

Signup provides ten free credits, with Lite and Pro plans above; amounts are not disclosed on the page we read. API access is stated for Pro and team plans, and commercial use is confined to Pro.

That last restriction is the one to notice before anything else. Commercial use gated to the top tier means the cheaper plan is effectively for experimentation only — and if you generate something on Lite and use it in client work, you are outside the terms you agreed to.

The free credits are a genuine trial rather than a teaser, which is to the site's credit and lets you assess output quality before spending.

The refund policy is published alongside terms, privacy, cookie, DMCA and disclaimer pages, which is a more complete set than many resellers offer. Read the refund terms specifically, because prepaid credits for generative output are a recurring source of disputes across this category.

We do not publish amounts. The comparison that matters is against direct access to the same capability, since that is what you are paying a margin above.

Gemini Omni: Resolution, Length And Terms