What does H3 Max Live mean?
H3 Max Live refers to using fal's H3 Max Director endpoint for a continuous, directable generated video session. The model does not finish one clip and wait for a new request. It keeps producing audiovisual output while control messages change what should happen next.
The official model identifier is minimax/h3-max/director.
The realtime control loop
- The application negotiates a WebRTC session through fal.
- It sends an opening
configuremessage with a prompt, aspect ratio, resolution, memory setting, and prompt version. - Video and audio arrive as a continuous media stream.
- The application sends another
promptmessage while the stream is live. - Provider events report whether that prompt is pending, applied, or rejected.
This event loop is why a proper interface needs more than a prompt box. A user should be able to see connection state, remaining paid time, live transcript, buffer condition, and the version of each direction.
Can voice control the stream?
Yes, but fal receives text control messages. AIR Live uses Gemini Live as the
speech-director layer: microphone audio is streamed from the browser, Gemini
keeps conversational context, and an actionable request becomes an
update_scene function call. AIR Live validates that call and sends its
direction to H3 Max Director.
The two live connections have separate jobs:
microphone -> Gemini Live -> structured direction
|
v
browser <- fal WebRTC <- H3 Max DirectorIf speech is unavailable, typed direction still uses the same video path.
Is it infinite?
The underlying model is designed for continuous generation, but applications must still handle provider session limits, billing, disconnections, and context windows. AIR Live currently sells explicit 60-second sessions rather than promising an unbounded stream.
Start directing
Use the realtime AI video generator or read the fal H3 Max API integration guide. AIR Live is independent and not an official fal or MiniMax service.