Google’s Lumiere Generates AI Videos That Feel Too Real

Date: January 29, 2024

Google’s new video generation AI tool, Lumiere, is set to burn the streets of video production with output that’s too real to feel unreal.

Google quickly notices the upcoming trends and brings the latest tech to facilitate them. With the boom of AI, Google has never stopped bringing new updates powered by Artificial Intelligence capabilities that have a huge impact on the overall AI development stats. In its latest development, Google has launched Lumiere, a generative AI video production platform, to set an example of GenAI's capabilities. Lumiere can create stunning life-like videos from simple text prompts in just a few seconds with an impressive frame rate of up to 80 per second. It can also create a video in a single process instead of the traditional frame-by-frame stitching.

Is There No Other In The Market?

The video generation tools currently available in the GenAI space have the capability to generate 25 FPS videos, with Stable Video Diffusion being the top leader of them. Google’s introduction of Lumiere’s capabilities has set fire to the current GenAI video production benchmarks. Runway, another powerful AI video generation tool, was tested with the same prompts, and the results were uncanny.

How Does It Work?

It works by understanding your text prompt and creating a base image to work on. It then uses Google’s indigenous technology Space-Time-U-Net (STUNet) to figure out how and where the objects would move in space based on changes in time. This method helps the tool create a single video instead of multiple points stitched together. The latest AI tool also creates videos at a much smoother frame rate than ever. It can generate videos in 80 FPS as compared to the 25 FPS output provided by other AI video generators.

AI tools’ outputs have gone from a vision of the future to the generators of futuristic reality. It simply means that Artificial Intelligence is growing much faster than anticipated. The sizzle reel published by Google and a pre-print scientific paper reveals that it has entered a market pre-occupied by competitors like Runway, Stable Video Diffusion, or Meta’s Emu.

Regarding limitations, Google’s Lumiere faces the same challenges as the other GenAI tools. All the AI video generators are facing difficulty in portraying movement. The leading aspect of Google is that the videos generated have a much more life-like perception regarding movements. But the overall video still feels unreal enough to create a strong dilemma about what’s real and AI generated.