Gemini Omni 1.1 Flash

Gemini Omni 1.1 Flash is Google’s high-performance multimodal model for fast video generation and editing, with natively synchronized audio. Unlike Veo, it runs on Google’s Interactions API, which processes text, image, audio, and video together for more cohesive, controllable output.

Highlights

• Native audio: every video is generated with a synchronized audio track — dialogue, sound effects, and ambient sound
• Text-to-video: generate a video from a detailed text prompt
• Image-to-video: animate a starting image into a video
• Keyframe interpolation: provide a start and end frame and the model generates a smooth transition between them — great for camera orbits, zooms, timelapses, and seamless loops
• Reference-to-video: supply reference images to guide the subject and style of the video
• Flexible output: 360p (fast draft), 720p, 1080p, or 4K, in 16:9 or 9:16

Usage

Text-to-video

import replicate

output = replicate.run(
    "google/gemini-omni-1.1",
    input={
        "prompt": "A cinematic drone shot through misty pine mountains at sunrise, gentle wind sounds",
        "resolution": "720p",
        "aspect_ratio": "16:9"
    }
)
print(output)

Image-to-video

output = replicate.run(
    "google/gemini-omni-1.1",
    input={
        "prompt": "Bring this scene to life with gentle camera movement and ambient sound",
        "image": "https://example.com/your-image.jpg",
        "resolution": "720p"
    }
)

Keyframe interpolation

output = replicate.run(
    "google/gemini-omni-1.1",
    input={
        "prompt": "Smooth continuous timelapse from sunrise to sunset over the lake",
        "image": "https://example.com/sunrise.jpg",
        "last_frame": "https://example.com/sunset.jpg"
    }
)

Prompting tips

• Be descriptive: describe the subject, action, camera movement, lighting, and mood
• Prompt the audio: describe dialogue (in quotes), sound effects, and music for a richer track
• Control scenes: add “in a single continuous shot” or “no scene cuts” for one unbroken take
• Use negatives in the prompt: e.g. “no dialogue”, “no music” — the model reads these directly
• Time events: you can write things like “after 3 seconds, a bird flies in” or use [0-3s] ... [3-6s] ... timecodes

Notes

• Every generated video includes a SynthID watermark, invisible to viewers but detectable for provenance
• Content safety filters apply to both prompts and generated video
• English is fully supported; other languages may work but are less tested