Gemini Omni 1.2 Flash vs 1.1: What Changed

Google released Gemini Omni 1.2 Flash on November 5, 2026, ten weeks after Gemini Omni 1.1 Flash. The upgrade that changes most workflows is length: one generation can now run from 3 to 15 seconds instead of stopping at 10. Around it, Google improved how closely the model follows reference images, made it possible to keep editing a finished clip over several turns, smoothed out dialogue and lip-sync, and cut the Gemini API price at 720p by 30% to $0.07 per second. Both versions are now available in the NanoBananas studio, and Gemini Omni 1.2 Flash costs fewer credits per second than 1.1.

This comparison was updated on November 11, 2026, when Gemini Omni 1.2 Flash went live in the NanoBananas studio. Model specifications come from Google's launch post, the Google Cloud model page and the DeepMind model card, all linked in the sources section. Credit prices and studio limits for both versions come from the NanoBananas pricing table. Quality claims about 1.2 below are Google's and are labeled as such.

What changed in Gemini Omni 1.2 Flash

Clips up to 15 seconds, set to the second

1.1 offered fixed lengths up to 10 seconds. 1.2 accepts any whole number of seconds from 3 to 15 for text, image and reference generation, and for editing. First-and-last-frame generation is still capped at 10 seconds, and scene extension adds 3 to 10 seconds per step. In practice, 15 seconds is long enough for a complete social ad, a product explainer or a short dialogue scene in one take, so you no longer have to stitch two clips and hope the join is invisible.

Better reference adherence

You can combine character, product or style images, short motion clips of up to three seconds, and audio cues in one request. Google says faces, garments, packaging typography and brand colors now hold across camera angles from the first frame to the last. The Cloud documentation allows up to 10 images and 3 videos per prompt. Google's launch examples include a 15-second virtual try-on built from one model photo and front-and-back shots of three outfits.

Multi-turn editing

After a clip is generated, you can keep refining it in conversation: swap the product or background, change the lighting or weather, or switch the aspect ratio, without generating the whole clip again. This works across full 15-second clips. For ad teams this is the most practical change, because one approved clip can become a set of variants for different audiences and placements.

More natural speech and steadier cameras

Dialogue sounds more fluid, lip-sync is tighter, and scenes with several speakers are easier to direct. Google also reports fewer unwanted jump cuts during long camera moves such as FPV flythroughs, orbits and crash zooms, with steadier scale through the move.

Lower price and general availability

The Gemini API price at 720p dropped 30% to $0.07 per second. 1.2 is also the first Gemini Omni release that Google made generally available on Google Cloud, with Provisioned Throughput and a listed retirement date of November 5, 2027 or later.

Gemini Omni 1.2 Flash vs 1.1 Flash: side-by-side

The upper rows describe each model as Google ships it; the last two rows are what the NanoBananas studio offers. Google's full 1.2 limits (10 images, 3 videos) are wider than what our provider exposes today. Prices in dollars are Google's list prices, not NanoBananas credits.

How multi-turn editing changes the workflow

With 1.1, a clip that was right except for one detail had to be generated again from the prompt, and the new take could differ in ways you did not want: a different camera move, a different expression, a different background. 1.2 lets you keep the take and change only what is wrong.

1. Generate the base clip with the performance, camera move and timing you want.
2. Edit one thing per turn. For example: "Replace the background with an oak forest; keep both characters, their movement and the camera framing unchanged."
3. Branch variants from the approved clip. Swap the product color, the season or the language on the packaging for different markets.
4. Switch the aspect ratio last, once the content is final, to produce 16:9 and 9:16 versions of the same spot.

In the NanoBananas studio, choose Gemini Omni 1.2 Flash, switch to the reference mode and upload the clip you want to change (2 to 15 seconds). The output keeps the length of your clip, so the duration setting does not apply, and you pay 1.2's per-second rate for the seconds of the uploaded video. You can add up to 5 reference images in the same request, for example the new product you want swapped in.

The same keep-then-change wording that works for image editing works here. Our photo editing guide explains the pattern in detail.

Audio and speech: what improved and what is restricted

Both versions generate speech, music and sound effects together with the picture. Most of the 1.2 improvement is in dialogue: voices sound less mechanical, mouths match the words more closely, and when a scene has several speakers you can assign voices to specific characters.

• Changing what a person says is restricted. DeepMind's model card says 1.2 can alter speech when editing a video, and that Google is restricting this for now.
• Audio upload is not supported on Google Cloud at launch. The Cloud model page lists audio as an unsupported modality, even though the launch post mentions audio cues as a reference type. Do not plan a workflow around uploading your own voice track until this is clarified.
• Every output carries SynthID, an invisible watermark, and the model supports C2PA Content Credentials.

What the price cut means in practice

The headline cut is 30%, but the larger saving comes from the workflow Google recommends: draft at 360p or 720p, fix details with follow-up edits instead of fresh generations, and upscale only the take you keep. If a 10-second ad used to take four full attempts on 1.1, the same result on 1.2 might take one generation and two edits.

NanoBananas charges credits, not API seconds. On 1.2 the studio charges 5 credits per second at 720P, 6 at 1080P and 10 at 4K, and it always shows the exact quote before you generate. The next section has the full tables for both versions.

Using Gemini Omni 1.2 and 1.1 on NanoBananas

Both versions support the same four modes in the studio: text to video, animating a first frame, start and end frames, and a reference mode. Audio is generated with every clip. The differences are length, reference limits and price.

Gemini Omni 1.2 Flash

Clips of 4, 5, 6, 8, 10, 12 or 15 seconds. Start and end frames are limited to 10 seconds. The reference mode takes up to 6 images, or one video of 2 to 15 seconds to edit; an image counts as one slot and a video as two, with 7 slots in total.

Gemini Omni 1.1 Flash

Clips of 4, 6, 8 or 10 seconds. The reference mode accepts up to 7 images plus one reference video of 2 to 10 seconds.

When you add a reference video, the studio bills the output seconds plus the reference seconds at 70% of the normal rate. An 8-second 720P clip guided by a 6-second reference video costs 79 credits: 8 credits × 14 seconds × 0.7, rounded up.

1.2 is cheaper than 1.1 at every resolution, so the main reason to stay on 1.1 is a shot you have already tuned there. Draft at 720P and switch resolution only for the final take. Start on the Gemini Omni 1.2 Flash page, the Gemini Omni 1.1 Flash page or the video studio.

Three prompts to compare 1.1 and 1.2

These prompts target the three areas Google says improved. Run each one on both versions at 720P and 10 seconds (80 credits on 1.1, 50 on 1.2) and compare the results side by side. Judge the checkpoints, not your overall impression.

• Is the word "AURELIA" still spelled correctly when it comes back into view?
• Do the cap shape and bottle proportions stay the same through the orbit?
• Do the reflections move with the camera, or do they flicker?

• Does each line come from the right person, with matching lip movement?
• Do the two voices sound like two different people?
• Do faces and hands stay stable while they speak?

• Does the shot stay continuous, or does it jump cut partway through?
• Does the scale of the rocks and waves stay believable as the camera rises?
• Does the sound fade in step with the camera move?

Should you switch to 1.2 or use another model?

For a broader comparison of video model families, see Seedance vs Veo vs Wan.

Known limits of both versions

• Fast sports action, rapid script changes and complex physics are areas Google says it is still improving in 1.2.
• Two aspect ratios only: 16:9 and 9:16. Crop in an editor for square or 4:5 placements.
• First and last frame stays at 10 seconds on 1.2, even though other modes reach 15.
• Speech editing is restricted on 1.2, and audio input is not supported on Google Cloud at launch.
• Watermarking: all outputs carry an invisible SynthID watermark.

Sources

• Google: Gemini Omni 1.2 Flash launch post: 15-second clips, reference adherence, multi-turn editing, $0.07 per second at 720p, availability and known weak areas.
• Google Cloud: Gemini Omni 1.2 Flash model page: durations, resolutions, input limits, supported modalities and retirement date.
• Google DeepMind: Gemini Omni 1.2 Flash model card: distribution channels and the speech-editing restriction.
• NanoBananas video studio pricing table: all credit figures and studio limits for Gemini Omni 1.1 Flash and 1.2 Flash.

Frequently asked questions