Odyssey: 'Hollywood-grade' AI Video Generation Platform in the Making

So behind the scenes, many hardworking startups are quietly developing new technologies related to visual AI for storytelling and filmmaking. These dedicated teams are experimenting with innovative ideas to help filmmakers and storytellers bring their visions to life in exciting new ways. One such startup, Odyssey, is using its experience from the autonomous vehicle industry to create tools that could make Hollywood-level visual effects more accessible and easier to control.

A few days ago Jeff Hawke has tweeted about his work on Odyssey, the venture, apparently developing Hollywood-grade visual AI. He views this technology as the next step beyond text-to-video which will empower storytellers in creating extraordinary movies, TV shows, and video games.

The development approach at Odyssey has been significantly influenced by autonomous vehicles (AV). Jeff, along with Oliver Cameron and a substantial portion of their team from Cruise, Wayve, Waymo, and NVIDIA, has spent years advancing AV technology.

Why is AV relevant?

1. AV advanced the state-of-the-art (SOTA) in visual AI to become operational.
2. AV scaled performance using real-world data.
3. AV required scalable, physics-accurate simulation systems.
4. AVs are trained to navigate the world; Odyssey is trained to generate worlds.

After years of pushing the SOTA in visual AI and observing the advancements in generative ML, Jeff's team was confident that AI could produce Hollywood-grade outputs. The question then was: what abstractions would make this technology useful for movie-making and beyond?

Imagine if ChatGPT output text that couldn’t be edited; you wouldn’t use it. That’s the current state of text-to-video, where the model’s output is a fully-formed video file that cannot be altered. If any element such as scenery, characters, lighting, or motion is incorrect, you have to start over.

Given their AV experience, his thoughts naturally turned to 3D and generating layers of 3D worlds.

Odyssey Training

Hence, they are training models to generate:

1. 🧊 High-quality geometry
2. 🏔️ Photorealistic materials
3. 🌄 Stunning lighting
4. 🏃‍♀️ Controllable motion

With this approach, users can generate video at AI speed while maintaining fine-tuned control over all the crucial details.

Although they are early in their journey, the progress, Jeff writes, has been remarkable. They are eager to collaborate with Hollywood to shape the future. 

The startup is currently expanding their R&D team and looking for the right individuals, so check out their jobs section.

* * *

I'm excited to see where this is going and what will come out in the future. Controllable AI video generation would no doubt revolutionize this budding new industry. It’s an exciting time for visual storytelling, and I can't wait to see the amazing content that will be created with these new technologies.