Runway has launched Runway Ads, an AI system that generates video and image advertisements, publishes approved versions to Meta, Google and TikTok, reads back performance data and produces the next round of creative based on what earns spend. The product, announced September 30, packages the workflow Runway built for its own marketing team into a product now being piloted with select enterprise partners, including AI development platform Lovable as a launch partner.

The origin story is the pitch. Runway’s growth marketing and creative teams were struggling to produce enough ad variants to keep up with the campaigns they wanted to run, so they built an internal agent to do it. According to the company, the system lifted its weekly output from 77 active ads to roughly 900 across Meta and TikTok, nearly doubled return on ad spend, raised conversion rates about 34 percent and cut cost per subscriber 41 percent. Those figures describe Runway’s own marketing program rather than an independent customer study, so they read as a ceiling, not a promise.

The workflow runs generate, approve, publish and learn. The agent recreates existing ads, spins out variations, localizes them for other markets and publishes only the human-approved versions. Human approval is a built-in step rather than an afterthought, which matters for brands that cannot afford unreviewed content going live at scale. Brands remain responsible for platform compliance, such as labeling AI-generated content on Meta.

The launch lands as performance marketing teams face an explosion of surfaces and formats. Runway’s chief creative officer Jamie Umpherson framed it as a response to a universal pain point: the number of placements brands must cover keeps growing while creative resources do not. Connecting generation directly to publishing and measurement is the bet that generative video earns a durable role in campaign operations rather than remaining a novelty tool.

For creators who run ads for their own products, courses or client work, the practical move is to watch the pilot rather than wait for it. The underlying lesson is already usable: build a tight loop between creative variants and performance data, keep a human in the approval seat, and treat localization as part of the same process instead of a separate project. When wider access opens, the creators with that discipline in place will get the most out of it.