← back
The Next Medium: Why Real-Time Interactive Video Changes Everything — Ahmed Ahres, Reactor
Takeaway
Real-time control could turn generated video into a programmable, interactive medium.
Summary
- Reactor defines world models as real-time interactive video: an ongoing stream that users can change while it is being generated.
- Instant visual feedback gives creators control over generated scenes instead of repeatedly requesting fixed video clips.
- The talk separates continuously prompted video generation from Genie-like controllable worlds and illustrates changing a scene by prompting a cat to appear.
- Proposed applications include interactive entertainment, dynamically generated advertising, and controllable simulations for robotics training.
world-modelsvideo-generationinteractive-video
Original description
Uber could not exist without GPS. Ahmed Ahres uses that to argue real time is a change of medium rather than a speedup: before GPS you consulted a map somebody else had already made, and afterwards your own position became something you could act on continuously. He runs the same argument through film. Once a viewfinder showed you what you were shooting, you could adjust while shooting, and that is the reason Instagram and TikTok were possible at all. Generated video today sits on the wrong side of that line. You prompt, you wait, you get a file back, and there is nothing further to do with it. His definition of world models is not Gaussian splatting and not a longer clip, but video that is interactive, effectively infinite, and generated fast enough to steer, which he demonstrates by prompting a cat into a scene while that scene is still generating. What it unlocks divides three ways: control, which he sums up as instant feedback being the ultimate form of it; character driven worlds that reach past games into robotics training data and into education; and live avatars, which he is refreshingly candid are still not working properly. The engineering consequence is that none of the batch playbook carries over. You are streaming pixels rather than returning files, every session is stateful and has to remember what happened when a character looked away, and sub 100 millisecond latency means putting GPUs near users rather than in one region. Speaker info: https://x.com/Boudatw / ahmedahres https://www.ahmedahres.com Timestamps: 0:00 - World models, defined as real time interactive video 1:44 - What happens when video becomes programmable 3:25 - Maps to GPS, and film to viewfinder 5:04 - Model one: infinite, interactive, real time 5:52 - Control, advertising, and instant feedback 7:32 - Model two: controllable worlds, robotics, education 9:13 - Model three: live avatars, not cracked yet 10:06 - What people are actually building on it 12:37 - Why real time infrastructure is not batch infrastructure 16:10 - Evaluation is still an unsolved problem