Runway unveils Runway Models v3 with low-latency on-device inference

Tech · 4 min read

Runway Models v3 includes quantized and sparsified versions of their generative backbones so applications can run inference locally with minimal quality loss. The company provided SDKs for macOS, Windows, and Android, enabling designers to run image generation, semantic editing, and style transfer without round trips to the cloud.

Runway highlighted use cases such as collaborative in-studio editing, live prototyping, and offline-capable workshops. The release also included privacy-focused deployment options, allowing studios to keep sensitive assets entirely local while still using generative features.

Tool developers appreciated the lower integration costs and faster feedback loops, though some enterprise customers pointed out that on-device models require additional QA for varied hardware. Runway responded with recommended hardware profiles and a profiling tool to estimate performance across machines.