fal's real-time AI video bet: post-trained models and creative control

Key points

Key takeaways from an a16z conversation between General Partner Jennifer Li and fal co-founder Gorkem Yurtseven, with Head of Engineering Batuhan Taskaya (September ư2026):

Transcript unavailable — summary based on title and description only.

Real-time AI video. Generative video fast enough to run in real time changes how creators work, per fal's founders.

H3 Max. fal's post-trained version of MiniMax's open-weight video model cuts generation time sharply while maintaining quality, via model post-training combined with systems and hardware optimisation.

Continuous video. That speed has enabled experiments with streams that remember previous scenes and respond to new directions while running.

Next challenge: control. The next bottleneck may be less about speed and more about control — camera movement, lighting, characters, motion and lip sync.

Creative workflows. Artists and studios could get real-time, steerable generation in professional pipelines, changing how AI video fits production.

Read more: a16z — How Real-Time AI Video Is Changing How Creators Work