Open Models Spark Video Ecosystem
Cristóbal Valenzuela, CEO of Runway, on the state of generative AI in video
Open-sourcing Latent Diffusion mattered less as a direct business model and more as a way to turn one research project into the default building block for a huge creator ecosystem. Once the weights and code were public, developers could build interfaces, fine tunes, plug ins, and workflows on top of the same core image model, while Runway kept moving up the stack into faster inference, better tooling, and production video products where customers actually pay.
-
The clearest proof of community innovation is what followed. Latent Diffusion was released in late 2021 through Runway and LMU Munich, then became the technical base for Stable Diffusion in 2022, which sparked a wave of third party tools like AUTOMATIC1111 and later ComfyUI that made model use far easier and more flexible.
-
That split explains Runway’s strategy on cost. Open models help the ecosystem invent new use cases, but video products are won in deployment. Runway describes video as harder than text because each request involves frame consistency, encoding, streaming, and low latency interaction, so the real moat is the system around the model, not just the model itself.
-
The economics are also very different from chat. Runway’s current pricing still meters video generation in credits and seconds, with API credits priced at $0.01 each and consumer plans starting at $12 per month. That reflects a workload where a single creative action can mean generating or transforming many frames, not returning one short text response.
The next phase is a market where open models keep spreading creative techniques, while value concentrates in companies that can package those techniques into fast, reliable, collaborative video software. As video generation gets cheaper, the winners are likely to be the teams that control workflow, dataset access, and product distribution, not just the base model release.