Gemini Omni 1.2 Flash
Generate 360p or 720p video with native audio using the Gemini Omni 1.2 Flash API. Use text or up to six reference images for a 3–15 second clip, or set a first frame with an optional last frame. A clip with a last frame can run up to 10 seconds. Make a product demo from photos taken at different angles, or connect two storyboard images into a shot that moves through a shop entrance. Find video generation request examples, supported parameters, and usage pricing on the model page.
Overview for yir
Gemini Omni 1.2 Flash API Guide for Reference Images and Frame Transitions
Gemini Omni 1.2 Flash makes short videos with audio from written prompts, product photos, or storyboards. OpenAny offers text-to-video, image-to-video with a first frame and optional last frame, and video generation with multiple reference images. Use frames to set how a shot begins and ends, or use several references to guide the look of a product, setting, or material. Google DeepMind model card · OpenAny route parameters
Example output
Watch Google's Pantheon fly-through ↗
Watch Google's ramen shop opening and closing frame comparison ↗
Example | Input and action | What to check |
|---|---|---|
Pantheon fly-through | Two architectural references guide a camera through the plaza and bronze doors and up toward the dome. | Check whether the layout makes sense as the camera moves indoors and whether the shot feels continuous. |
Ramen shop transition | An exterior opening frame and an interior closing frame guide a forward move into the shop. | Check how the two frames connect and whether the furniture stays consistent during the move. |
Google published both demos. The ramen shop clip compares versions 1.1 and 1.2 side by side. The OpenAny quickstart below lets you try a similar continuous move. Google announcement and demos
Modes and settings
Mode | Input | Duration |
|---|---|---|
Text to video | Describe the subject, movement, camera path, and sound in one prompt. | 3–15 seconds |
First and optional last frame to video | Supply one | 3–15 seconds with only a first frame, or 3–10 seconds with a last frame |
Reference images to video | Supply one to six images with the | 3–15 seconds |
All three OpenAny modes offer 360p or 720p, in 16:9 or 9:16, with one clip per task. Use publicly accessible HTTPS URLs for images.
Quickstart
In the OpenAny Playground, choose Text to video. This prompt generates one continuous shot from the street into a bookstore. Send a POST request to https://api.openany.si/v1/videos with the header Authorization: Bearer $OPENANY_API_KEY.
{
"model": "gemini-omni-1.2-flash",
"route": { "provider": "yir" },
"route_mode": "balanced",
"input": {
"type": "text",
"prompt": "One continuous eye-level dolly shot from a rain-wet sidewalk through the open door of a small bookstore. Pass a row of shelves and stop on a warm reading lamp beside an armchair. Maintain a steady walking pace and consistent lighting transitions. Include distant rain fading into quiet indoor ambience, with no dialogue."
},
"parameters": {
"duration": 8,
"resolution": "720p",
"aspect_ratio": "16:9",
"n": 1
}
}
Poll GET /v1/videos/{task_id} and retrieve the video URL from the result's file list when the task succeeds. Check the layout around the doorway, the change in lighting, and whether the camera comes to rest at the reading nook. If you have an exterior image and an interior image, try the first-and-last-frame mode and keep the clip within 10 seconds.
Generation and editing
Google also demonstrates conversational editing of previously generated clips. The OpenAny modes covered here generate new clips through the asynchronous video API. Google's native API capabilities