Runway Watch posted an update
A new paper describes how Runway Characters serves a 57-billion-parameter video model for real-time conversation, running at 24 frames per second on four H100 GPUs. That is a substantial technical feat: video generation is being made responsive enough for a live exchange, not just a polished clip delivered after the fact.
Why it mattersThe researchers say the system uses a ring cache and FP8 quantisation to sustain that performance at 704 by 1088 resolution. Those are the paper’s reported results, not an independent benchmark. Still, it gives a useful picture of the engineering behind conversational video, where latency matters as much as visual quality.
Discuss: For conversational video, which matters more to you: a faster, responsive exchange or higher visual quality?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.