Runway launches router for generative media models
Runway's new Media Router picks a video, image or audio model for each API request based on preferences users set for cost, quality and latency, so builders no longer have to choose models by hand.
Runway announced Media Router, a new feature in its Runway Dev platform. For each request, it chooses which video, image or audio model will do the generation. Runway calls it the world's first model router for generative media, which brings the kind of routing already common for LLMs to media generation.
Runway Dev offers Runway's own models, including Gen-4.5, Aleph 2.0 and Act-Two, along with third-party models such as Seedance, GPT Image 2 and ElevenLabs. Runway says customers have struggled to keep up as new models arrive and prices change, and that teams often stay with a model long after it has stopped being the best choice.
Setup has three steps. In the Runway Dev portal, users save a named config that ranks cost, quality and latency, sets a price cap, and lists which providers or models are allowed or blocked. Requests then go to a single endpoint with that config ID attached instead of a model name. The router first drops any model that can't handle the request, falls on a block list or costs more than the cap, which is set per second of video or per image. It then scores the remaining models against the user's preferences and runs the request on the top one. The response says which model was used and why.
If no model meets the constraints, the router returns an error that names the constraint that ruled everything out, rather than quietly relaxing the settings. A dry-run flag shows which model would be picked, at no cost and without generating anything.
Media Router is live now for video, image and audio models. Runway has not said what routing costs or how it measures quality when scoring models.