Gemini Omni 1.1 Flash gives developers real control over generated video

Generating video has so far stumbled on the fact that the result is a black box: you write a prompt and hope a usable clip drops out. Google DeepMind is now releasing Gemini Omni 1.1 Flash into the API, which changes this.

The end of random cuts and visual chaos

The update allows native extensions to existing video in ten-second blocks, up to a maximum length of 40 seconds. The model holds the context from the previous scene, so the camera doesn't jump elsewhere and the visuals remain consistent. This makes it a real storytelling tool, not just a random b-roll generator.

The testing phase gets cheaper for creators

Another shift is support for rough 360p previews. This saves a third of the cost and significantly speeds up iteration when looking for the right camera movement before the developer pays for the final render.

4K output is nice, but smoothness is paramount

The model can work with the specification of the first and last frame for smooth transitions and looping, and finally pull the resolution up to 4K. While the market is catching up easily on resolution, an API for real manipulation of movement is what most models have been missing.

The video API market shifts toward production

What will determine Omni's success is the stability of its API and its price compared to competitors (Runway, Luma). If it maintains visual continuity, it will become the standard for applications that want to integrate video, not just show it in a demo.

Lilith's verdict

AI video finally gets what will make it a tool and not a toy. Anyone who just offers a 'generate' button will soon be unusable for enterprise customers.