Runway previews real-time AI video model that generates frames in under 100 milliseconds
Runway and NVIDIA just previewed a real-time HD video model that generates frames in under 100 milliseconds.
Runway unveiled a research preview of a new real-time video generation model developed in collaboration with NVIDIA today. Showcased at the GPU Technology Conference (GTC), the system drastically cuts the time-to-first-frame down to under 100 milliseconds for high-definition video.
Right now, AI video generation is a notoriously slow process. Users typically sit in server queues waiting minutes for a single short clip to render. With Runway’s sub-100-millisecond latency, the generation process becomes nearly instant. This turns prompt-to-video into a live, interactive stream rather than a batch rendering task.
Here is what powers the new capabilities:
- The hardware: The model runs on NVIDIA’s Vera Rubin architecture. This next-generation AI accelerator platform is built specifically for massive video inference.
- The foundation: The real-time integration strengthens Runway’s General World Model (GWM-1). This system simulates physics, geometry and lighting across interactive environments.
Still, there is a massive catch. The research preview currently relies on enterprise-scale server racks estimated at nearly $4 million each. That staggering cost puts the technology completely out of reach for regular users.
On the flip side, this hardware is not a permanent consumer bottleneck. The co-design between Runway and NVIDIA is engineered to drive down the long-term price of generation as the hardware scales.
- The efficiency gains: NVIDIA’s Vera Rubin architecture delivers up to 10 times better inference throughput per watt. That design drastically reduces the effective cost-per-token.
- The cloud rollout: Cloud providers will adopt this new hardware and fractionalize the capacity. This allows creators to access real-time generation through standard API subscriptions.
The Bottom Line: The picture here is clear. Running a real-time video model on bleeding-edge hardware is currently a massive barrier to consumer access. However, as hardware scales the software will follow. That progression will eventually bring generation times down to seconds for everyday users. We all knew this was coming, but this demo is simply that exact expectation unraveling in real time.
If you need on-demand GPUs for training, fine-tuning, inference, or running open-source models, give RunPod a try.
- Available hardware: H100, H200, A100, L40S, RTX 4090, RTX 5090, and 30+ more
- Cost: significantly cheaper than AWS or GCP, billed per second, no contracts
- Setup: spins up in under a minute, 30+ regions worldwide
Get the core business tech news delivered straight to your inbox. We track AI, automation, SaaS, and cybersecurity so you don't have to.
Just read what you want, and be done with it.
