London-based AI lab Odyssey has launched a analysis preview of a mannequin remodeling video into interactive worlds. Initially specializing in world fashions for movie and sport manufacturing, the Odyssey staff has stumbled onto probably a very new leisure medium.
The interactive video generated by Odyssey’s AI mannequin responds to inputs in real-time. You’ll be able to work together with it utilizing your keyboard, cellphone, controller, or finally even voice instructions. The oldsters at Odyssey are billing it as an “early model of the Holodeck.”
The underlying AI can generate realistic-looking video frames each 40 milliseconds. Which means whenever you press a button or make a gesture, the video responds nearly immediately—creating the phantasm that you just’re truly influencing this digital world.
“The expertise in the present day seems like exploring a glitchy dream—uncooked, unstable, however undeniably new,” based on Odyssey. We’re not speaking about polished, AAA-game high quality visuals right here, not less than not but.
Not your normal video tech
Let’s get a bit technical for a second. What makes this AI-generated interactive video tech totally different from, say, a normal online game or CGI? All of it comes all the way down to one thing Odyssey calls a “world mannequin.”
In contrast to conventional video fashions that generate total clips in a single go, world fashions work frame-by-frame to foretell what ought to come subsequent primarily based on the present state and any person inputs. It’s just like how giant language fashions predict the following phrase in a sequence, however infinitely extra complicated as a result of we’re speaking about high-resolution video frames quite than phrases.
“A world mannequin is, at its core, an action-conditioned dynamics mannequin,” as Odyssey places it. Every time you work together, the mannequin takes the present state, your motion, and the historical past of what’s occurred, then generates the following video body accordingly.
The result’s one thing that feels extra natural and unpredictable than a standard sport. There’s no pre-programmed logic saying “if a participant does X, then Y occurs”—as an alternative, the AI is making its greatest guess at what ought to occur subsequent primarily based on what it’s realized from watching numerous movies.
Odyssey tackles historic challenges with AI-generated video
Constructing one thing like this isn’t precisely a stroll within the park. One of many largest hurdles with AI-generated interactive video is maintaining it steady over time. While you’re producing every body primarily based on earlier ones, small errors can compound rapidly (a phenomenon AI researchers name “drift.”)
To deal with this, Odyssey has used what they time period a “slender distribution mannequin”—primarily pre-training their AI on basic video footage, then fine-tuning it on a smaller set of environments. This trade-off means much less selection however higher stability so every little thing doesn’t change into a weird mess.
The corporate says they’re already making “quick progress” on their next-gen mannequin, which apparently reveals “a richer vary of pixels, dynamics, and actions.”
Working all this fancy AI tech in real-time isn’t low cost. At present, the infrastructure powering this expertise prices between £0.80-£1.60 (1-2) per user-hour, counting on clusters of H100 GPUs scattered throughout the US and EU.
That may sound costly for streaming video, however it’s remarkably low cost in comparison with producing conventional sport or movie content material. And Odyssey expects these prices to tumble additional as fashions change into extra environment friendly.
Interactive video: The subsequent storytelling medium?
All through historical past, new applied sciences have given start to new types of storytelling—from cave work to books, pictures, radio, movie, and video video games. Odyssey believes AI-generated interactive video is the following step on this evolution.
In the event that they’re proper, we may be wanting on the prototype of one thing that may remodel leisure, training, promoting, and extra. Think about coaching movies the place you possibly can observe the talents being taught, or journey experiences the place you possibly can discover locations out of your couch.
The analysis preview obtainable now’s clearly only a small step in direction of this imaginative and prescient and extra of a proof of idea than a completed product. Nonetheless, it’s an intriguing glimpse at what may be doable when AI-generated worlds change into interactive playgrounds quite than simply passive experiences.
You can provide the analysis preview a attempt here.
See additionally: Telegram and xAI forge Grok AI deal
Wish to be taught extra about AI and massive knowledge from trade leaders? Take a look at AI & Big Data Expo happening in Amsterdam, California, and London. The great occasion is co-located with different main occasions together with Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.
Discover different upcoming enterprise know-how occasions and webinars powered by TechForge here.
