[AINews] Fal’s H3 Max Live breaks the infinite videogen barrier
Source
latent.space
Date
Context
H3 Max Live illustrates how generated video can become an ongoing interaction: viewers influence what the stream produces next. In its September 17 technical account, fal describes keeping a connection open so generated audio and video flow to viewers while new prompts travel back to the running model. The company says audience-selected prompts change its fal.live broadcasts within seconds. The Director API documentation likewise supports prompt updates within an existing session. This provides a mechanism for audience-directed streams; the session design itself does not establish indefinite uptime or sustained story quality.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
