Odyssey, an artificial intelligence (AI) startup founded last year, revealed details of its first AI product on Monday. The company revealed that it is building an AI video model that can create Hollywood-quality visual effects, similar to OpenAI's Sora tool, which the company has yet to release. Odyssey's co-founder said the AI model will give users fine-grained editing and control over the output, adding that the company uses multiple large language models (LLMs) to generate different layers of the output video that can be controlled individually.
How Odyssey’s AI visual model works
In a series of posts on X (formerly Twitter), Odyssey CEO and co-founder Oliver Cameron said the AI startup has raised $9 million (approximately Rs 751 crore) in a seed round of funding led by Google Ventures and is currently building tools to deliver high-quality videos that users can customise and edit.
Cameron also shared details about Odyssey's AI technology, claiming it's designed to create “Hollywood-caliber” videos, and said the company has trained four generative models to give users “full control over the core layers of visual storytelling.”
Each model can be used separately to accurately compose every detail of your scene.
These models are then combined to produce the desired video or scene.
Additionally, our model output will be integrated into existing pipelines used in Hollywood and elsewhere. pic.twitter.com/jHZoevLV9n
— Oliver Cameron (@olivercameron) July 8, 2024
Cameron pointed out that a problem with existing AI text-to-video conversion models is the lack of tools and options to control or edit the output. “As a storyteller, you have little ability to dictate the environment or characters, or iterate on the details of a shot until you get it just right. We need more powerful models,” he added.
To solve this problem, the company uses multiple AI models to generate a single layer of composite video. Cameron says four models independently generate geometry, materials, lighting, and motion. These four layers are generated simultaneously based on a single text prompt, then combined to show the final video.
The company says users will have the option to configure each layer individually for greater control over the output, and Odyssey will also integrate its video output into existing Hollywood tools and systems used to produce visual effects.
Currently, the AI video model is in the early stages of development. There is no set release date for the AI model; however, the company emphasizes that it will continue to share regular updates on its progress. Of note, Cameron previously worked at Cruise and Voyage, two startups working on self-driving cars.
The company's other co-founder and CTO, Jeff Hawk, previously served as vice president of technology at Wayve, an AI company developing self-driving systems.
