Google launches AI video generator Gemini Omni 1.1 Flash

Google releases artificial intelligence model Gemini Omni 1.1 Flash Memorydesigned specifically for video generation. The project attempts to solve the problem of how to generate organic video from a given sample of multiple frames – giving developers more control over what happens between the first and last frames.

Image source: blog.google

The most important innovation of Google Gemini Omni 1.1 Flash is the ability to remember existing video when the model is instructed to continue a scene; the model relies not only on the last frame, but also on the previous 10 seconds of video footage. This helps her keep the characters, setting, and overall direction of the scene more consistent without having to guess what’s going to happen next. The range of possibilities for continuing the original scenario has been expanded. The video can be extended in 10-second increments until the total length of the video reaches 40 seconds. This gap may not seem that long compared to traditional film, but it gives artificial intelligence a wider opportunity to produce sequences with a given beginning, middle, and end.

This means the user can provide the first and last frames and Google Gemini Omni 1.1 Flash will produce everything needed to connect them. This could be the movement of the camera around an object, smooth zooming of a composition, or looping a video. You can send up to three seconds of real-life footage to the model for additional visual context, such as how she herself will depict the character’s appearance in the generated scene. The pre-production phase doesn’t require the highest quality footage – they can be generated in 360p, allowing AI to run 60% faster than 720p for a third of the cost. You can test ideas, tweak cues, try multiple versions, and when the results are optimal, Omni 1.1 Flash will create a 1080p video or scale it to 4K.