Gemini Omni 1.2 Flash in Givon: 15-second clips, references and Russian dialogue

One reference image and four videos, all in 16:9 at 720p with sound. The three Omni 1.2 videos are first takes. For the Omni 1.1 product example, we used a second take after the first produced a defect in the tin. The video prompts and reference image were unchanged. We made the tin image with the built-in imagegen tool, adding only a request for a single photorealistic 16:9 image to the original prompt.

We generated the 1.1 clip through UseAPI → Google Flow and the three 1.2 clips through OpenRouter, selecting Gemini Omni 1.2 Flash explicitly. These are different provider routes, so four examples are not a controlled benchmark of the models. The cover is a frame from the 15-second clip below.

We asked for a morning scene in an apartment, a close-up of coffee and a balcony finale, all built around the same product.

The first take includes all three shots: the woman opens the tin, coffee pours into a cup, and she drinks it on the balcony. The final shot ends with a camera pullback. Her hairstyle changes between the apartment and balcony scenes, though, so the result needs a closer check if character continuity is essential.

One coffee tin with Cyrillic lettering, one reference image and the same prompt for both versions. We looked at the large brand name, which means “Quiet Morning,” as well as the packaging’s colour and shape.

Both versions kept the recognisable green cylindrical tin. The large lettering is easy to read in 1.2, but the hand barely turns the tin. In the new 1.1 take, the tin is lifted and tilted, and the large lettering remains legible. This example does not establish that 1.2 preserves text better through a full rotation. Check the small print in the original video before using it in an ad, too.

We asked for a café conversation with distinct voices, no music and no subtitles. The English prompt contains the original Russian lines.

The first take includes both requested lines. In English, they mean “Are you really leaving tomorrow?” and “Just for a week, don’t worry.” We also checked the words with local speech recognition, without giving it the expected text. This is one successful Russian-language example; we did not separately measure lip-sync accuracy.

In 1.2, you can generate from text, a first frame or reference images, or create a transition between a first and last frame. Reference images cannot be combined with specified frames. The image limit depends on the provider route: up to six through OpenRouter or seven through Google Flow.

You can also edit a source clip up to 15 seconds long, or extend a video shorter than 30 seconds by adding 3–10 seconds. Both modes require an MP4 file up to 100 MB. Extension supports 360p and 720p. Editing and extension use Google Flow, so their availability depends on the provider account’s Google Flow plan and access to version 1.2.

Editing is charged for the length of the source clip, rounded up to the next whole second, with a one-second minimum. Extension is charged only for the seconds you add.

Studio members get 50% off Omni 1.2. The exact price, including your discount, appears before you start. Omni 1.1 keeps its own settings and prices.

Google reports a roughly 30% reduction in its 720p API price. That compares Google’s API rates; it does not mean a 30% price cut in Givon, where the cost depends on the available provider route and video settings. Google Flow produces 1080p and 4K by upscaling the base video.