Gemini Omni Flash Review: Google's New #1?

Google DeepMind announced Gemini Omni on May 19, 2026 at I/O. The pitch is "create anything from any input, starting with video." You can mix text, images, audio and video as inputs, then change the result in plain language over several turns. On August 27, 2026, Google released Gemini Omni 1.1 Flash as a generally available API model. That release added scene extension, first and last frame control, cheaper 360p previews and upscaling to 4K.

This review is based on Google's documentation, two public leaderboards and published hands-on reports. We did not run our own generations for it. Every number links to its source, and no vendor pays for placement on PromptVideo.

Gemini Omni Flash 1.1 at a Glance

What it costs per second

Google publishes a price per token, not per clip. The Gemini Enterprise Agent Platform pricing page lists the output tokens for each second of video with audio: 1,931 at 360p, 5,792 at 720p, 8,688 at 1080p and 17,376 at 4K. At $17.50 per million output tokens, that comes to roughly:

Input tokens cost extra, at $1.50 per million (pricing page), but they're small next to output. The 360p tier is the one to use for iteration. Google says 360p previews come back up to 60% faster at a third of the cost of 720p (Google). That lets you test prompts cheaply and pay for 720p only on the takes you keep.

Is Gemini Omni the New #1?

It's #1 for Google, and #1 on one major leaderboard. It isn't #1 across the board.

On Arena, the original Gemini Omni Flash is #2 on text-to-video at 1513 with 26,576 votes, so Google holds the top two spots. The 1.1 model has only 1,784 votes, though, and Arena gives its rank a spread of 1 to 4. FLUX 3 Video (1493), Grok Imagine Video 1.5 (1492) and Seedance 2.0 (1479) are close behind. Artificial Analysis shows a different picture. Wan 3.0, MiniMax H3, Seedance 2.5, FLUX 3 Max and HeyGen Video 1 all rank above Omni there, and Artificial Analysis puts its likely rank between 9th and 11th.

The one consistent result is that Omni beats Veo 3.1 on all four boards, usually by a wide margin. Google is acting on that. The Gemini app page says "Gemini Omni will replace Veo in the Gemini app", and the Gemini API video docs say the Veo 3.1 Preview models were shut down on October 22, 2026, with gemini-omni-1.1-flash as the recommended migration path.

Tested: What Public Reports Show

None of these results are ours. They come from published hands-on reports and the model card, and each one is linked.

JXP sells access to AI models, so it has a commercial interest. Its numbers come from a small sample of the first Omni Flash release, before 1.1.

Strengths

• Editing by conversation. You can generate a clip and then ask for changes, such as swapping an object, restyling the shot or changing the camera, without starting over. The Gemini app lists video-to-video editing and multi-turn editing as core features. The Verge reported that, unlike Veo, Omni can take an existing video as its starting point (The Verge).
• World knowledge. Google told The Verge the model has "a lot" more world knowledge than Veo, which helps prompts that name real places, objects or situations.
• Price at 720p. About $0.10 a second puts it well below Veo 3.1 Standard and Seedance 2.5 at list price (see the comparison below). Artificial Analysis lists it at $9.12 per minute against $24 for Veo 3.1.
• Cheap drafts. Native 360p at a third of the 720p price is uncommon among top-ranked models.
• Reach. It's in the Gemini app, Flow, YouTube Shorts and the Gemini API, and Google names Adobe Firefly, Figma Weave, Runway and GMI Cloud as partners building with it (Google).
• Social formats. Clips of 10 seconds or less in 9:16 with native audio match what social media and TikTok need, which is why our homepage picks it for those formats.

Weaknesses

• Short single takes. 10 seconds per generation. Extension reaches 40 seconds, but each 10-second step is a new generation, so you get more chances for drift. Seedance 2.5 and Wan 3.0 produce up to 30 seconds in one pass (dbggr comparison).
• No native 1080p. 1080p and 4K come from upscaling 720p output, which isn't the same as generating at that resolution.
• Two aspect ratios. 16:9 and 9:16 only, with no 1:1 or 4:5 for feed ads.
• Edit drift and text. Google's own model card names consistency across edits, complex motion and accurate text as open problems.
• Image-to-video ranking. On Artificial Analysis it's 14th for image-to-video. If you animate product shots or stills, Seedance 2.5 and MiniMax H3 rank higher.
• Restrictions. The model card says edits that change what a person says are restricted. JXP reported that brand names in prompts were blocked.

How It Compares to Veo, Sora, Kling, Seedance and Runway

List prices below come from the dbggr AI video comparison (October 13, 2026) unless noted. Ranks are from Artificial Analysis text-to-video and Arena text-to-video.

• vs Veo 3.1: Omni ranks higher on every board, costs less than Veo 3.1 Standard, and has replaced Veo in the Gemini app and the Gemini API preview. Veo's advantage is longer extension at 720p. JXP's comparison still rated Veo 3.1 stronger for cinematic fidelity.
• vs Sora: OpenAI discontinued Sora. The app closed in April and the API shut down on September 24, 2026. See our Sora alternatives guide.
• vs Kling 3.0: Kling is a little cheaper per second, offers longer 15-second clips and goes up to 4K. Omni scores higher on both Artificial Analysis boards.
• vs Seedance 2.5: Seedance ranks higher on Artificial Analysis and produces 30-second single takes, but costs more than twice as much per second at 720p. Our Seedance 2.5 review covers where that's worth paying for.
• vs Runway: Runway Gen-4.5 ranks far lower on Arena, and Runway is one of the companies building on Omni (Google).

Who It's For

• Good fit: short-form social video, TikTok and Shorts creators, teams that iterate a lot and want cheap 360p drafts, and anyone who wants to fix a clip by asking for changes instead of rewriting the prompt.
• Good fit: developers already using the Gemini API who need a replacement for the retired Veo 3.1 Preview models.
• Poor fit: short films and long UGC ads that need 15 to 30 seconds in one take, square feed ads, and anything with a lot of on-screen text.

How to Access Gemini Omni

1. Gemini app: Requires a Google AI Plus, Pro or Ultra subscription, and you must be 18 or older. Features vary by plan and region (Gemini).
2. Google Flow: Omni 1.1 Flash is in Flow for AI Plus, Pro and Ultra subscribers. In The Verge's hands-on, a generation cost 15 to 40 Flow credits and each round of edits 40. The $20-a-month AI Pro plan included 1,000 credits a month, and about 20 clips left the reviewer with 145 (The Verge).
3. YouTube Shorts and YouTube Create: free, per Google's launch post.
4. Gemini API and Google AI Studio: use the model code gemini-omni-1.1-flash. The older gemini-omni-flash-preview has been shut down (model page).
5. Gemini Enterprise Agent Platform: for Google Cloud billing and enterprise controls.

If you're comparing subscription bundles that include several models, our Higgsfield pricing breakdown shows how credit plans compare with paying for the API directly.

Frequently Asked Questions

What is Gemini Omni?

Gemini Omni is Google DeepMind's model for generating and editing media from any mix of inputs, starting with video. It takes text, images, audio and video and returns video with native audio. You can then edit the result through conversation (Google).

Is Gemini Omni Flash the best AI video model?

Only on some leaderboards. It's #1 on Arena's text-to-video board (September 21 snapshot), but 10th on Artificial Analysis text-to-video, where Wan 3.0 leads. It is clearly Google's best video model and ranks above Veo 3.1 on every board we checked.

How much does Gemini Omni cost?

In the API, it costs about $0.10 per second at 720p, according to Google's pricing page. By our math, that's about $0.034 at 360p and $0.30 at 4K. There's no free API tier. In the apps it comes with Google AI Plus, Pro and Ultra subscriptions and is free in YouTube Shorts.

How long can Gemini Omni videos be?

A single generation runs from 3 to 10 seconds. With scene extension you can add 10 seconds at a time up to 40 seconds total (Google).

Does Gemini Omni replace Veo?

Yes, in Google's own products. Google says Omni will replace Veo in the Gemini app, and the Veo 3.1 Preview models were shut down in the Gemini API on October 22, 2026, with Omni as the suggested replacement (Gemini API docs).

Gemini Omni vs Kling 3.0: which should I use?

Use Kling 3.0 when you need clips longer than 10 seconds or native high resolution. Use Omni when you want higher benchmark scores, editing through conversation and cheap 360p drafts. Kling is slightly cheaper per second at 720p.

Can you tell if a video was made with Gemini Omni?

Yes. Google says it adds a SynthID watermark to every Omni video. That matters for disclosure, because The Verge found its likeness clips convincing.

What's the best way to try it?

Start in AI Studio or Flow with 360p drafts, keep edits to about four turns, and render at 720p only once the shot works. For help picking a model for a specific format, use the picks on the PromptVideo homepage or browse resources.