AI Video Just Became Programmable Television

A strange thing happened when fal made H3 Max faster than realtime. The obvious benchmark story was: a 5-second video can render in under 3 seconds. The more interesting story is what developers started building once generation became faster than playback. They built television. The first demo got banned Fal engineer Rehan Sheikh connected H3 Max to a livestream inspired by the "interdimensional cable" idea from Rick and Morty. The important technical property was simple: generation time < playback time If the system can finish the next segment before the current one ends, the stream does not need to stop. The experiment went viral. It was removed from Twitch and then Kick. Rehan later said both platforms banned the stream within an hour. Instead of continuing to move between platforms, fal built its own. fal.live is interactive AI television fal.live is fal's experimental streaming platform. It currently exposes generated channels such as: Chaos Sitcom Anime Soap Opera Popcorn Availability is still experimental, and individual channels can pause or reconnect. The interesting part is the interaction model. Viewers are not just watching. They influence what happens next through prompts and voting. That turns the stream into a feedback loop: audience -> direction -> model -> generated scene -> audience reaction -> next direction 📺 That loop is much closer to a game engine than a traditional video file. Infinite Slop makes the idea even clearer Pieter Levels (@levelsio) built Infinite Slop with fal. The site describes itself very accurately: an endless AI-generated TV channel People type what they want to see next. Suggestions enter a queue. The system generates new scenes continuously. It is chaotic and often absurd, but I think the roughness makes the underlying shift easier to see. The audience is moving from spectator to input device. H3 Max Director is the real technical story The important release underneath these demos is H3 Max Director. A normal video generation endpoint behaves roughly like this: prompt -> request -> finished clip Director behaves more like this: open session -> start stream -> send direction -> continue scene -> send another direction -> continue again The session is carried over WebRTC. According to fal's documentation, each generated segment continues with 39 frames of carried context. Director also retains previous direction prompts as memory: 12 by default, configurable up to 50. That matters because continuity is not being approximated only by giving the next request the previous final frame. The model keeps a running context and evolves the same stream. Fal's public Director sessions currently run up to 2 minutes. Longer sessions are being enabled gradually for approved use cases. This changes what a video API can be Most video APIs today produce assets. You request something and receive a file. Director starts looking more like a runtime. That enables a different class of product. Interactive stories A viewer can choose what a character does next while the story is already playing. Generative games Game state can be converted into directions for the model. Live visual agents Speech, sensor data, an LLM or any other source of text can steer the visual stream. Adaptive marketing A live experience can react to the audience rather than playing one prerecorded creative. Synthetic channels Instead of storing a fixed content library, the channel continuously produces the next segment. This is why I think "AI video generation" is starting to become too narrow a phrase. The interesting system is not a generator. It is an interactive visual process. The cost is already calculable Director currently bills by generated video time. During its launch discount: $0.02 / second After the promotion: $0.08 / second If I naively kept a Director feed generating 24/7 at the current promotional rate: $1,728 / day $51,840 / 30 days $630,720 / year At the $0.08/sec list price: $6,912 / day $207,360 / 30 days $2,522,880 / year That is obviously not a casual hobby stream. But I also do not think the right comparison is with a YouTube subscription. The comparison is with the cost of producing, operating and continuously supplying content for a real channel or interactive visual experience. From that perspective, the economics are already interesting. This is the worst version of this medium fal.live and Infinite Slop are rough. That is exactly why I am paying attention. The first websites looked bad. The first streamed video looked bad. The first mobile games looked primitive. The first useful version of a medium usually arrives before the polished version. H3 Max Director gives developers a new primitive: a visual world that can keep running and accept new instructions while it runs. ⚡ That sounds much more consequential than another "best video model" leaderboard result. 2026 might be remembered as the year generative video stopped being a file and started becoming a process.