Its over now ...

Yours truly has long waxed poetic about writing, graphics and video, concepts thought to remain the sole province of man forever but no more as these disciplines, and others, are now being impacted by AI in ways thought impossible to contemplate just two years ago. When seeing Lumiere's short clips, the future of video comes into view, ushering in a an era of film making open to all as long as one's computer's connected to the net. How good will the quality be? Well, it will vary depending on the person creating the video but wait, as stated in BRT, the next step is to enable AI to drive said video as that's coming whether we like it or not.

I saw an excellent presentation on the power of Chat GPT a week ago. Excellent to a fault save that the prediction of AI becoming sentient is definitely 10 years off without question. IMHO, making a blanket prediction such as this is not viable because there's no certitude thanks to Quantum Mechanics. Shit always happens. Just ask Murphy to see why. :)

In this "wonderful" blog, I stated Google's resources in building a truly intelligent AI smarter than Chat GPT's is coming true because now, Deep Mind's tech is being folded into Google's Gemini as Deep Mind's years long research into science, medicine and game theory will help transform Google's AI into AGI (Artificial General Intelligence) within a guesstimate of 3-5 years. Regarding ASI (Artificial Super Intelligence),  just add another two or three as it will now become a scaling issue and not a research one. Additionally, we haven't even discussed the notion of analog computing, the "old" tech promising to change everything in terms of just how truly pervasive AI will become as civilization moves further into the 21st century.

The Tech

We introduce Lumiere -- a text-to-video diffusion model designed for synthesizing videos that portray realistic, diverse and coherent motion -- a pivotal challenge in video synthesis. To this end, we introduce a Space-Time U-Net architecture that generates the entire temporal duration of the video at once, through a single pass in the model. This is in contrast to existing video models which synthesize distant keyframes followed by temporal super-resolution -- an approach that inherently makes global temporal consistency difficult to achieve. By deploying both spatial and (importantly) temporal down- and up-sampling and leveraging a pre-trained text-to-image diffusion model, our model learns to directly generate a full-frame-rate, low-resolution video by processing it in multiple space-time scales. We demonstrate state-of-the-art text-to-video generation results, and show that our design easily facilitates a wide range of content creation tasks and video editing applications, including image-to-video, video inpainting, and stylized generation.

 

No doubt, as a videographer, I will be using this tech when it goes live.

Regarding Its over now ...