Reading time: 7 min
In short. On 30 September 2026, Runway, best known for its generative video models, announced Runway Ads: an agent that designs ads, publishes them on Google, Meta and TikTok, reads the results and produces the next round. For advertisers, the promise is compelling: test far more creative without multiplying production costs. Runway’s headline numbers are striking, but they come from its own campaigns, with no published methodology. Since 2 August 2026, the EU AI Act also imposes transparency obligations on generated content. Our agency view: a serious tool for creative production at scale, provided you keep control of the brand, the measurement and the account access.
What is Runway Ads?
Runway describes Runway Ads as an “autonomous performance marketing engine”. It is built on Runway Agent, the platform’s creative assistant, which gained marketing briefs, performance analysis and campaign and localisation skills over summer 2026. According to Runway co-founder and co-CEO Anastasis Germanidis, the company built it because it needed it for its own marketing. The difference from a plain image generator: Runway Ads connects to ad accounts and closes the whole loop, from creation to delivery to learning.
Runway says it has run the system on its own paid marketing since July 2026. First reserved for enterprise partners, Runway Ads became generally available on 27 October 2026, included in every paid Runway plan, with subscriptions, purchases, leads, app installs and brand awareness as goals. It appears in paid accounts as a separate “Ads” workspace, with a dashboard, campaigns, a review queue, an asset library and brand kits. Dedicated apps complete the set, such as “Vary Ad”, which produces four variations of an existing ad from a simple instruction (headline, background, palette). Runway has not published Runway Ads-specific pricing or a list of supported countries.
The create, publish, learn loop
The workflow described by Runway follows six steps, repeated continuously:
- Context intake: brand guidelines, product imagery, catalogue feeds for e-commerce, past ads and campaign instructions.
- Variant generation in image and video: hooks, layouts, copy, and automatic resizing to each channel’s formats (vertical, square, landscape).
- Automated brand checks before the approval queue.
- Human approval, on by default. Auto-publishing can be enabled campaign by campaign or by variant type.
- Publishing to the connected Google Ads, Meta and TikTok accounts.
- Reading results and generating new variants from the ads that win the most spend.
Guardrails are built in: a cap on daily budget changes, a minimum share of new audiences, retargeting limits, and localisation rules per market (tone, terms never to translate, on-screen text). These settings matter as much as visual quality: they are what stops an agent from optimising a short-term metric at the expense of the brand.
The claimed results, and how to read them
Runway backs its launch with results from its own campaigns since July 2026:
| Metric reported by Runway | Claimed result | Our reading |
|---|---|---|
| New ads per week | From 77 to about 900 | The volume gain is the most credible: it is the tool’s core function. |
| Return on ad spend (ROAS) | Doubled | Period, budget and attribution model not disclosed. |
| Conversion rate | About +34%, click-through rate flat | No documented control group. |
| Cost per subscriber | −41%, despite higher spend | Measured on a consumer subscription product; hard to transpose to B2B or luxury. |
These results should not be dismissed, but they are not proof. An advertiser who multiplies creative volume tenfold often improves results whatever the tool, simply because bidding algorithms get better inputs. The only measurement that counts is a controlled test on your own accounts, against a reference group.
Google, Meta, TikTok: what is actually covered
Runway Ads produces images and video. It naturally fits where creative diversity drives performance, much less in text ads on search engines.
| Platform | Relevant uses | Watch-outs |
|---|---|---|
| Meta (Facebook, Instagram) | Reels, Stories and feed variants, hook testing | The tool’s most natural ground; “AI info” label to manage. |
| TikTok | Native vertical video, market-by-market versions | AIGC label required for realistic generated content. |
| Google Ads | Image and video assets for Performance Max, Demand Gen and YouTube | Search text ads remain outside the tool’s core. |
A practical point we noticed when connecting: when a Google Ads account is linked through a manager account, the tool sees every linked account. An agency should therefore restrict access to the relevant account only, or obtain each client’s written consent before connecting their accounts to a third-party tool.
A market converging on creative automation
Runway is not alone. The ad platforms themselves are building creative generation into their tools:
- Google unveiled Gemini Omni video creation in Asset Studio and one-click creative testing at Google Marketing Live on 20 May 2026, first for Performance Max and Demand Gen.
- Meta, according to trade press, aims for fully automated ad creation from a product image and a budget, building on Advantage+.
- TikTok offers its Symphony suite to generate and adapt video ads.
Runway’s edge is its position: an independent video studio able to drive all three platforms from one workspace and one brand kit. For multi-channel brands, that matters. The flip side is dependence on a third party that aggregates performance data, creative and account access.
AI Act, platform labels and data: the rules that apply
Article 50 of the EU AI Act has applied since 2 August 2026. It requires providers of generative systems to mark their outputs in a machine-readable format, and those who publish deepfakes to disclose them. The “digital omnibus” Regulation (EU) 2026/1744 gives systems placed on the market before 2 August 2026 until 2 December 2026 to apply that marking. In practice, a photorealistic AI-generated ad must be identifiable as such.
Platforms add their own rules: Meta’s “AI info” label, TikTok’s AIGC label, and YouTube’s altered or synthetic content disclosure for realistic scenes. On data, connecting a third-party tool to a client’s ad accounts is a matter of contract and GDPR: a written mandate, minimal access scope and an audit trail of actions.
Premium and luxury brands: the risk of commoditisation
An agent optimising cost per conversion will mechanically favour whatever hooks work best in the short term: discounts, urgency, the visual codes of online retail. For a luxury house or a premium brand, that is exactly what to avoid. As we argued in our analysis of luxury branding and generative AI (FR) (in French), AI creates value when it is directed, and destroys value when it is used to mass-produce without an artistic eye.
For these brands we recommend keeping human approval on every variant, locking the brand kit (palette, typefaces, framing, banned words) and reserving the tool for variations and testing, while the master creative remains the work of an art director.
Our test protocol
Rather than comment on the promises, we are testing Runway Ads on Million Marketing’s own campaigns before offering it to clients. The protocol:
- Duration: two to three weeks, on Performance Max, then on Meta.
- Capped budget and no auto-publishing: every variant is approved.
- Brand kit locked in the tool before the first generation.
- Metrics: cost per qualified lead, conversion rate, perceived creative quality against our AI Studio work, and production time saved.
We will update this article with the results, including if they disappoint.
What this changes for agencies
If producing variants becomes almost free, the value of a paid media agency shifts to what the agent does not do: cross-channel budget allocation, conversion signal quality, art direction, compliance and the business reading of results. Agencies that sold production volume will face competition. Those that sell judgement and measurement become the natural pilots of these tools.
Frequently asked questions
What is Runway Ads?
An advertising agent launched by Runway on 30 September 2026. It generates ad images and videos from a brand kit, publishes them on Google, Meta and TikTok after approval, then creates new variants based on the results.
Does Runway Ads publish ads without approval?
No. The launch announcement described human approval as on by default, with auto-publishing possible per campaign. The general availability announcement of 27 October 2026 states that every ad requires team approval before it publishes.
Do AI-generated ads need to be disclosed?
For realistic content, yes: Article 50 of the AI Act has applied since 2 August 2026, and Meta, TikTok and YouTube enforce their own labels. Minor retouching does not require it.
Does Runway Ads replace an agency?
It replaces part of variant production. It replaces neither strategy, nor art direction, nor measurement. For a premium brand, it is the combination of both that delivers results without damaging the image.
Sources and references
- Runway, Introducing Runway Ads, 30 September 2026
- Runway, Runway Ads Is Generally Available, 27 October 2026
- Runway Help Center, Vary Ad
- Runtime Wire, Runway launches an AI ad platform built on its own campaign workflow
- AlphaSignal, Runway Ads automates Meta, Google and TikTok campaigns
- Fullstack Creators, Runway Agent 2.0 wants to be your ad agency
- EUR-Lex, Regulation (EU) 2024/1689 (AI Act)
- ActuIA, AI Act: obligations in force and delays under omnibus 2026/1744
- Gibson Dunn, EU AI Act Omnibus Agreement
- PPC Land, Asset Studio gets Gemini Omni and 1-click creative testing at GML 2026
- Campaign Asia, Meta aims to fully automate ad creation with AI by 2026
- Cinerads, AI ad disclosure requirements in 2026
Read next
On the same topic
AI Act and AI-Generated Advertising Content: What Article 50 Changes for Agencies and Large Accounts in 2026Read the article →