Google Opens Up Gemini Omni 1.2 Flash: Up to 15-Second Video at $0.07 per Second

If you have ever made a video with a generative model, you know the main pain point: it came out almost right, but the background is off, the lighting is cold, and the character suddenly changed faces midway through the clip. You have to regenerate everything from scratch and pay again. With the Gemini Omni 1.2 Flash update, Google targets exactly this pain while also cutting the price.

What has changed

The key numbers first. The maximum clip length has grown from 10 to 15 seconds, and you can set the length in one-second steps, from 3 to 15. In Google's A/B tests, version 1.2 beats 1.1: 63% to 37% on reference-based generation (reference-to-video) and 59% to 41% on image animation (img2vid). This is the company's own data, so treat it as a guideline rather than an independent benchmark.

The second important novelty: multi-step editing in chat. You get a clip and, with the next prompt, can change the background, lighting, weather or aspect ratio without rebuilding the whole video. In essence, this is a move from "generate and pray" to a familiar cycle of revisions.

Third: references. You can upload up to 6 images (3 per character, up to 2 characters) and video references of up to 3 seconds. Faces, clothing, packaging and colors should be preserved throughout the frame. For advertising and e-commerce this is critical: product packaging must not "drift." In addition, there is native audio, lip sync and multiple speaking characters, plus first and last frame control for transitions between scenes (up to 10 seconds only).

How much it costs

The price for 720p is $0.07 per second, so a 15-second clip costs about $1.04. Version 1.1 cost $0.10 per second, so the savings are around 30%. The other rates:

• 360p: $0.023 per second
• 1080p: $0.104 per second
• 4K: $0.209 per second

Hence a practical workflow: run drafts at 360p, and once a scene is approved, upscale to 1080p or 4K. Ten 15-second iterations at 360p cost about $3.45, which is less than two final clips in 4K. For a product manager this means you can budget for many revisions rather than one expensive attempt.

Where to try it and what's wrong

The model is available in AI Studio and the Gemini API, in Flow, in Google Vids (where 15-second clips come at no extra charge with a regular Google account), in the Gemini app for AI Ultra subscribers, and in Google Ads. The model itself has no free tier, only the paid API, so you will need to attach a card to experiment.

Google itself acknowledges the limitations: the model still struggles with fast sports, abrupt scene changes and complex physics. If your use case is a boxing match or falling objects, budget time for manually picking the good takes. In addition, an imperceptible SynthID watermark remains in the output, so the origin of a clip cannot be hidden. For most companies this is more of a plus, but those who want "clean" material should take it into account.

What to do in practice

Developers building video features into products: rework your pipeline around a "360p draft, then upscale" scheme. It is the simplest way to cut costs several times over. Product managers and marketers: test references with packaging and brand faces, the main scenario for ad previews. Founders: the price drop from $0.10 to $0.07 per second and clips of up to 15 seconds make the unit economics of video services noticeably more attractive. But before going to production, run the model on your own hardest scenes: A/B test figures speak to the average, and your clips may land squarely in the problem zone.