Gemini Omni 1.2 Flash Reference to Video

Gemini Omni 1.2 Flash Reference to Video

What is Gemini Omni 1.2 Flash Reference to Video?

Gemini Omni 1.2 Flash Reference to Video is Google's reference-guided AI video model. Give it 1 to 10 reference images and a prompt, and Gemini Omni 1.2 Flash builds a new clip that keeps your characters, products and brand marks consistent, with native synchronized audio generated in the same pass. Unlike an image-to-video pass that animates one first frame, this model treats each image as a reference to carry into a scene you describe, so a product shot on a plain backdrop can appear on a rotating podium, or a stylised mascot can act out a fresh shot. It is the fast member of the Gemini Omni family: draft at 360p, then render the winning take up to 4K. Clips run 3 to 10 seconds in 16:9 or 9:16.

Key Features

• •Reference-to-video from 1 to 10 images with product, character and brand-detail consistency.
• •Native synchronized audio generated with the picture.
• •Legible on-screen text, spelled as written.
• •360p, 720p, 1080p or 4K output with a 360p draft-and-upscale workflow.
• •Strong on stylised looks: motion-graphics, claymation, low-poly and 3D product shots.
• •SynthID watermarking and C2PA Content Credentials on every clip.

Best Use Cases

This model is strongest where a known subject has to stay on-model: product-feature shots with on-screen callouts, brand explainers, animated infographics, and stylised character pieces. The 360p draft mode makes Gemini Omni 1.2 Flash a fast way to storyboard a reference-driven idea before a 4K render.

Prompt Tips and Output Quality

Bind each reference with its tag (IMAGE_REF_0, IMAGE_REF_1, ...) and name what to keep: colour, shape, logo. Describe a fresh environment, not the reference backdrop. Quote on-screen words and ask for them spelled as given; order beats "first, then, finally"; name a stylised medium; use a locked-off camera and list what must not move; write one sound per beat. Roughly one instruction in a dense prompt may be dropped, so check the referenced subject and any text before upscaling.

How to use Gemini Omni 1.2 Flash Reference to Video

Gemini Omni 1.2 Flash Reference to Video builds a new clip from 1 to 10 reference images and a prompt, keeping the referenced subjects consistent while placing them in a scene you describe. Reach for this model when a specific product, character or brand mark has to stay on-model across the shot, with native audio generated in the same pass.

Prepare your references

Upload 1 to 10 clear, well-composed stills. Each reference is addressed in the prompt by its tag in order: IMAGE_REF_0 for the first image, IMAGE_REF_1 for the second, and so on. Use separate references for a product and its logo, or for two subjects that must share one scene.

Write the prompt

• •Bind each reference and say what to keep, for example "the speaker from IMAGE_REF_0, keep its colour, grille and carry-loop".
• •Describe a fresh environment; this is reference-to-video, not a first-frame animation.
• •Quote on-screen words exactly and end with "every word spelled exactly as given, no other text".
• •Order beats with "first, then, finally", name a stylised medium, and write one sound per beat.
• •Use a locked-off camera and list what must not move for clean layouts.

Recommended Settings

• •duration: 5 seconds for a single beat; up to 10 for a short sequence.
• •resolution: 360p to draft cheaply, then 720p or higher; legible on-screen text needs 720p or more, and finals can go to 1080p or 4K.
• •aspect_ratio: 16:9 for landscape, 9:16 for mobile and social.
• •seed: leave at -1 for variety; set a fixed integer to reproduce a take.

In a tested 720p five-second run, two references, a product and a brand emblem, carried cleanly into a new studio scene with correctly spelled callouts and a matched audio track. About one instruction in a dense prompt can be dropped, so review the referenced subject and any text before upscaling.

How many reference images can I use?

One to ten per request.

Does Gemini Omni 1.2 Flash generate audio?

Yes, native synchronized audio on every clip.

How long can a clip be?

Three to ten seconds.

What resolutions are supported?

360p, 720p, 1080p and 4K, in 16:9 or 9:16.

Is the output watermarked?

Yes, SynthID plus C2PA Content Credentials.