📰 Key Takeaways

Gemini has launched the Omni 1.1 Flash version, adding a suite of creative control and generative video features, made available to developers through the Gemini API in Google AI Studio. The goal is to give generative video tools greater controllability, faster iteration speed, and quality that meets the bar for production deployment.

This update centers on three core features. The first is “Scene Extension,” which seamlessly continues an existing video from its ending point. Omni 1.1 can analyze up to 10 seconds of prior footage as reference — a big jump from the older model, which could only reference the last 1 second — resulting in much better visual continuity and narrative consistency. Each extension adds up to 10 seconds, and you can chain extensions up to a total of 40 seconds. Developers can implement this through the Gemini API’s interactions.create method, passing the previous_interaction_id parameter to chain onto the previous video segment.

The second feature is “start and end frame control,” letting users set the starting and ending frames of a shot separately, with the model generating a coherent transition between the two keyframes. This is well suited for complex camera moves, zoom transitions, or seamless looping clips.

The third is a 360p draft mode. Compared to Omni 1.1’s standard 720p resolution, 360p generation can be up to 60% faster and costs just a third as much — ideal for rapid prototyping, storyboard iteration, and fast preview rendering on developer platforms.

💬 JudyAI Lab Take

Gemini’s Omni 1.1 Flash rollout, opened up to developers through the Gemini API in Google AI Studio, is really about pushing generative video’s controllability and iteration speed up to production-deployment standards — not just staying at the demo-reel level.

What’s worth noting for AI builders is the shared thinking behind these three features: Scene Extension stretches the reference window from 1 second to 10 seconds (with up to 40 seconds of cumulative extension), start/end frame control makes camera moves and transitions predictable, and the 360p draft mode trades resolution for much cheaper, faster iteration. This reflects generative video tools moving from “one-shot output” toward a “controllable, iterative” workflow — developers aren’t just tossing in a prompt and waiting for the result anymore; they can work more like editing a film, chaining, adjusting, and iterating quickly. For teams integrating generative AI into their products, this “cheap draft first, high-quality final pass later” layered approach is worth studying.

If you’re evaluating a generative video API, it’s worth first getting clear on what you actually need: one-shot quality, or the flexibility to iterate repeatedly.

📅 Source Info

🔗 Further Reading