Fun leads to everything Can we make a scavenger hunt fun while interleaving it with skits powered by the most advanced video-to-video AI workflows?
VIDEO
Time pressure is no stranger to film production. Time is everything. So a fast stream-lined game begins to feel like an efficient distributed production environment. Right?
Try completing a scene:
Scan to record on your phone. (Or click if you are on it already) You need to film a horizontal clip, about 4-10 seconds long. Then generate (optionally) with a director's note about what your scene should look like. Every scene comes with: a location, a character, a voice and a line of dialog. Actually you can say whatever you want. But if you go off script and ruin the story beat, then its likely to be rejected by the story owner.
Last weekend we hosted a Scavenger Hunt at Ai On the Lot. I worked with Jay Judah, Adam Mutchler, and the extended network of Machine Cinema in Los Angeles. At the largest Ai media tech conference, we allowed attendees to inject their performance into a sample story. What story you ask? Jay Judah asks us to imagine a temporal rift has opened causing Culver City's film history to spill into the present - an experimental short exploring how we might interact with the media of the past.
A few weeks before the conference 10zebra built out the generative workflows, the game experience, and the admin portal to oversee and create new scavenger hunts. Through a series of tests with volunteers and across the core team, we iterated on capture, omni model providers, system prompts and live pre-viz. There were many awkward videos of 1930s police officers running around snowy streets, with the voice of a young girl. Voice morphing, stuttering, character blending. All hurdles to overcome. Even from a month ago, we have come a long way. But compare it to our pilot experiments with GAN models many years ago? Night and day.
"The future that we dreamed of and built for - it's finally here." - Philip Meier, Founder of 10zebra During the scavenger hunt we could watch our live portal of the script being converted into community generated video. For example: every time someone filmed a new version of Scene 4, it replaced the previous version of THE PIANIST in the SPEAKEASY and everyone could watch it on the next loop. The reel always looped back to the beginning, so as more of the story completed, the draft narrative grow longer and longer. Eventually our live draft covered the script. Last weekend we set a modest goal: 10 scenes, with 10 lines of dialog. Short enough that a team could complete all scenes, and long enough to test some core questions.
Watch the live feed:
Can different people act out different lines, and still look like the same character? Yes.
Can they also sound like the same character? Yes.
Can we capture nuance of their pacing, and intonation? Yes.
Can we capture their postural motions and transfer that to the generated scene? Mostly. Sometimes. It depends what action they do, and a complimentary prompt really helps . We call them directors notes, and they help your make believe actions land as believable in the narrative world, vs looking like awkward movement. How are they used? Directors notes are instructions that are "interpreted" by the software and provided as auxiliary information to a multimodal "Omni" model. We don't build these models, but we support all the major ones. Our contextual management makes it easier to manage your acting with respect to a story.
If you have not noticed it yet, there is a coming tsunami of AI generated content, and we foresee that audiences will identify with craft behind the work. Automated content saturates. Software without effort lacks struggle and the layering of human intention. How you made it - and what it's about - become increasingly important, above and beyond what it looks like. At 10zebra we believe this means engaging meaningfully with actors, values-based brands, fans and community members.
So - on to the next step - lets build an ecosystem where we can reliably pay creators to contribute performances - while protecting their likeness from being abused by contextual use outside the project. Ultimately this will come down to contracts with production companies (our clients), but we are going to try to pave the path for scenarios that work for everyone - from talent to brand.
%2C%20Color%3DNegative.svg){kind=icon}
