Four resolutions, priced apart
360p at $0.03 a second for drafts, 720p at $0.10 as the native render, 1080p at $0.15 and 4K at $0.30 as upscales of it. Iterate cheap, finish sharp.
Model · Google Gemini Omni 1.1 Flash
Gemini Omni 1.1 Flash — the world model, now with room to grow
The second generation of Google's Omni video model, in production here since September 2026. Same always-on sound and physics-aware motion as before, plus the things people kept asking for: a 360p draft mode at a third of the price, 1080p and 4K output, a closing frame to land on, up to three video references, and scene extension that keeps appending until the clip is 40 seconds long. $0.10 a second at 720p, paid with crypto.
01What it is
What is Gemini Omni 1.1 Flash?
Gemini Omni 1.1 Flash is the second generation of Google's Gemini Omni family — a world model that reasons about a scene the way a physics engine would and renders it as video with synchronized sound. Google announced it in August 2026 and shipped it to the API as gemini-omni-1.1-flash-preview. Against the first generation it adds four resolutions instead of one (360p, 720p, 1080p, 4K), a last frame for interpolation, video clips as references, and a scene-extension task that reads up to ten seconds of what came before and continues it. On BananaBanana it is the model behind Omni Flash in the studio and the MCP server; the original generation stays available as Omni Flash 1.0 for anyone who wants it.
What changed from Omni Flash 1.0?
Mostly the edges of the envelope. The first generation was a single 720p tier: 3–10 seconds, one clip, no way past ten seconds except cutting. 1.1 keeps the 3–10 second unit but lets you draft at 360p — up to 60% faster and a third of the price — then re-run the winner at 720p or push it to 1080p and 4K. It accepts a last frame next to the first one and animates between them. It takes up to three short video references to carry a character or a style across clips. And it can extend a scene: hand it a finished clip and it appends 3–10 more seconds, again and again, until the whole thing is 40 seconds. Google says the model now looks at up to ten seconds of prior footage while extending, where the old one only saw the last second — which is why the continuation below does not drift.
02Capabilities
What Omni 1.1 Flash can do
Scene extension to 40 s
Pick a finished clip, describe what happens next, and the model appends 3–10 seconds to the end of it. Repeat until the clip is 40 seconds. The source is read in full every time.
First and last frame
A starting image and, optionally, a closing one: the model interpolates the motion between them. Reveal shots, orbits and product turns become predictable.
Video references
Up to three clips of three seconds each, alongside up to ten images. They carry a character, an outfit or a style into a brand-new scene without a starting frame.
Always-on synchronized audio
Every clip ships with generated sound — ambience, effects, speech. Steer it with an "Audio:" line in the prompt; there is no separate switch.
Conversational and video-to-video editing
Describe a change and the scene regenerates with character, lighting and camera intact. Any finished video — a Veo clip, your own upload — can be sent in for a restyle at the same length.
360p draft · $0.09
4K · $0.90
03Limitations
What it still can't do — honestly
These are limits of Google's API, not of our integration, and they matter before you spend a dollar. 1080p and 4K are upscales of the 720p render, not more detail — worth it for delivery, wasted on drafts. Extension only appends: nothing can be inserted at the start or in the middle, and an uploaded clip that already contains speech cannot be given new dialogue. A single clip is still 3–10 seconds; the 40 seconds come from chaining extensions, and each extension is billed on the full resulting length, so the last step of a 40-second chain costs as much as a 40-second clip. Video references are trimmed to three seconds each, and an edit still returns exactly the length of its source.
Also missing from the API: seeds that reproduce a composition, a negative-prompt field (write "no…" in the prompt instead), multiple samples per request, and switching the audio off. In the EU, Switzerland and the UK Google does not allow uploaded videos with people to be edited or extended. If one of those is a dealbreaker, the Veo 3.1 family has seeds, negatives, silent clips and 4K native rendering, and Wan 3.0 renders 30 seconds in one take.
04How it works
From draft to 40 seconds in three passes
Draft at 360p
Write the prompt with an "Audio:" line, pick 16:9 or 9:16 and the exact length, and render at 360p — $0.24 for eight seconds. Fix the composition here.
Render the keeper
Re-run the winning prompt at 720p ($0.80 for eight seconds), or at 1080p or 4K if it is going on a big screen. Add a last frame if the shot has to land on something.
Extend the scene
Open the clip from the feed, choose Extend, say what happens next. Each pass appends 3–10 seconds and the chain stops at 40. Reference images can bring a new subject into the continuation.
Reference image
Extended clip · 8 s + 3 s
05Pricing
Priced by the second, by resolution
| 模型 | 3 s | 8 s | 10 s | Sound |
|---|---|---|---|---|
| Gemini Omni 1.1 Flash · 360p$0.03/s — drafts, up to 60% faster | $0.09 | $0.24 | $0.30 | Always on |
| Gemini Omni 1.1 Flash · 720p$0.10/s — the native render | $0.30 | $0.80 | $1.00 | Always on |
| Gemini Omni 1.1 Flash · 1080p$0.15/s — upscaled | $0.45 | $1.20 | $1.50 | Always on |
| Gemini Omni 1.1 Flash · 4K$0.30/s — upscaled | $0.90 | $2.40 | $3.00 | Always on |
| Gemini Omni Flash 1.0 · 720pthe previous generation, 720p | — | $0.80 | $1.00 | Always on |
| Veo 3.1 Fast · 720pGoogle's model, 4–8 s only | — | $1.00 | — | Costs extra |
Prices are per finished clip and include the audio track. A dash means the model cannot render that length. An edit is a new generation of the same length as its source; an extension is billed on the full resulting length (source plus the appended part). Full tables on the pricing page.
Compare every model on the full pricing page, or read what the first generation still does well on the Omni Flash 1.0 page.
06Learn more
Guides from the blog
Gemini Omni Flash API Review: What the Preview Really Does
Hands-on review of the Interactions API — what it accepts, what it refuses, real per-clip costs.
阅读指南AI Image to Video: Turn Any Photo into a Moving Clip
First frames, end frames, loops and extended videos across the models here.
阅读指南Veo 3.1 Prompt Guide: Write Prompts That Actually Work
Camera language, lighting, motion — the structure carries over to Omni almost unchanged.
阅读指南07FAQ
Gemini Omni 1.1 Flash, answered
What is Gemini Omni 1.1 Flash?
The second generation of Google's Gemini Omni world model for video, released in August 2026 as gemini-omni-1.1-flash-preview. It generates 3–10 second clips with synchronized sound at 360p, 720p, 1080p or 4K, animates between a first and a last frame, takes video clips as references, and extends a scene in 3–10 second steps up to 40 seconds. On BananaBanana it is the model behind Omni Flash.
How much does a Gemini Omni 1.1 Flash video cost?
By the second and by resolution: $0.03 at 360p, $0.10 at 720p, $0.15 at 1080p and $0.30 at 4K. So a 3-second draft is $0.09, a 10-second 720p clip is $1.00, and ten seconds of 4K is $3.00. An extension is billed on the full resulting length. You top up with crypto from $1 and pay per clip; failed generations are refunded.
阅读完整解答How does scene extension work, and how long can a video be?
You pick a finished Omni clip — your own upload works too — choose Extend, and describe what happens next. The model reads the whole source, appends 3 to 10 seconds to its end, and returns one continuous file. Run it again on the result and the chain grows until it reaches 40 seconds. Reference images can bring a new subject into the continuation; nothing can be inserted before or in the middle of the existing footage.
阅读完整解答Is 4K real, or upscaled?
Upscaled. The model renders natively at 720p; 1080p and 4K are upscales of that render, and 360p is a smaller, cheaper draft render. The 4K file really is 3840×2160 and looks clean on a large display, but it does not contain more detail than the 720p pass. Veo 3.1 is the model here that renders 4K natively.
What are the 360p drafts for?
For deciding before you pay. A 360p clip costs a third of 720p and comes back up to 60% faster, so you can try three phrasings of a shot for the price of one final. Once the composition, camera and sound are right, re-run the same prompt at the resolution you deliver in. The draft and the final are separate generations, so expect a different take, not a pixel-identical upscale.
Should I use Omni 1.1 Flash, Omni Flash 1.0 or Veo 3.1?
Omni 1.1 Flash for anything with sound that needs to grow past ten seconds, land on a specific frame, or reuse a character from a clip. Omni Flash 1.0 if you specifically want the previous generation's take — it is the same price at 720p and still edits conversationally. Veo 3.1 when you need seeds, negative prompts, silent clips or native 4K, and Wan 3.0 for a single 30-second take.
Do I need a Google account or credit card?
No. BananaBanana accepts crypto only — USDT, USDC, BTC, ETH, SOL, TON and other major coins, minimum top-up $1. Sign up with an email, top up, generate; unused balance never expires, and new accounts start with $0.20.