Connect the Ad Account, and Leave Approval On

Runway Ads will generate the variant, run a brand check, and wait. Leave the wait on.

On 30 September 2026, Runway published the announcement. Connect an ad account and a brand kit, and the product generates video and image creative, publishes approved variants to Meta, Google, and TikTok, reads performance back from those platforms, and produces the next round from what earned spend.[1] The loop is closed. The word doing the work is "approved."

Human approval is on by default. An automated brand check runs before a variant reaches that queue. A team can move to automated publishing later, by campaign or by variant type, as they build confidence.[1]

If you are writing the brief, write the gate before you write the volume.

The page is a brief, not a click-path

I fetched the announcement on 1 October 2026. It is a product post. It does not name a menu, a toggle, or a "step 3, click Approve." If someone on your team asks for the click-path, the honest answer is that Runway has not published one. What the page does name is specific enough to brief a person, and specific enough to refuse a vague "just connect it."

The inputs are an ad account and a brand kit. Generation uses brand guidelines, past ads, and product imagery.[1]

The outputs are video and image ads. Each variant is resized and reformatted across aspect ratios and channel specs. Localization follows rules your team sets. The examples on the page are brand voice, and terms you do not want translated. That pass covers on-screen text and product screenshots, along with the voiceover.[1]

Spend moves inside limits you set. The three named on the page are a maximum daily change, a minimum net-new audience share, and retargeting caps.[1]

Then the sequence. Brand check. Approval queue. Publish to Meta, Google, and TikTok. Performance comes back. The next round is built around what earned spend.[1]

"We built Runway Ads because we needed it ourselves," Runway co-CEO Anastasis Germanidis says on the page. The line after that is the design claim: generation, publishing, and measurement in one system, "rather than three tools stitched together."[1] Treat it as their reason for building the product. Do not file it as a result.

The page also reports figures from Runway's own ads since July. I am leaving the numbers out. They are the company's account of its own program, not an independent benchmark, and not your number.

The door is a pilot

Access is the limitation, and it is a short one. Runway says it is piloting with select enterprise partners. The page is a form for Enterprise Sales. It is not a public switch you can flip tonight.[1]

Earlier paragraphs talk about small teams, small businesses, and individual creators as if the loop is already open to them. Read that as ambition. Brief from the access paragraph. That paragraph is a pilot, and a sales form.

You can still do the useful work. The decisions the page names do not require a login. Write them down. If the pilot opens, you connect the account to a brief that already has rules. If it does not, you still have a creative-ops brief you can hand a designer, a media buyer, or the tool you already use.

Write the kit before you connect the account

Start with the kit. The page treats it as an input, not a mood.

• Write the brand guidelines you would hand a new designer this week. Voice, claims you are allowed to make, product names that have to be spelled one way. A link to a deck nobody has opened is not a kit.
• Choose the past ads. The page lists past ads as an input. It does not say the system will sort the winners from the ones you pulled. Pick a small set that earned spend, and leave the rest out of the folder.
• Gather the product imagery. If you sell from a feed, the page's ecommerce example generates ads from existing product feeds rather than one listing at a time.[1] Use that path only if the feed is already clean. A messy feed publishes the mess, at the volume the page is selling.

Then write the rules the system is allowed to apply without stopping to ask you.

• Localization. Put the brand voice in a few lines a contractor could follow. List the terms that stay in the original language. Say, in the same note, that the rule covers the script, the on-screen text, and the product screenshots. The page is explicit about that coverage.[1] A term you protected in the voiceover can still land wrong on the pack shot if the rule never mentioned the pack shot.
• Spend fences. Set a maximum daily change. Set a minimum share of spend that stays on net-new audience. Set a retargeting cap. Those are the three limits the page names.[1] If you cannot put a number on one of them, you are not ready to let a system shift spend, even with a person still on the queue.
• The approval rule. Name the person who clears the queue. Start with every campaign, and every variant type, requiring that person. The page says automated publishing is a later choice, scoped by campaign or by variant type, after confidence.[1] Write what confidence means before anyone treats it as a feeling in a kickoff. A usable version: a variant type can leave the queue only after that person has approved a run of it, and the brand check has stopped flagging the same miss.

That last bullet is the how-to. The kit exists so the queue is reviewable. The fences exist so a yes in the queue cannot move the budget further than you already allowed.

A brand check is not an approval

The page separates two checks. They are not the same job.

The automated brand check runs before a variant reaches the queue.[1] The post does not say what the check scores. Do not fill that gap. Do not brief it as "it catches logos, claims, and legal" unless Runway has told you that in writing, on a page you can cite. What you know from this announcement is the order. Check first. Then a person.

Human approval is on by default.[1] Leave it there. Generation tools have been making variants for a while. This one publishes approved variants directly to Meta, Google, and TikTok, then reads performance back and makes the next round from what earned spend.[1] Once that loop is closed, a yes is not a file in a folder. It is spend.

The page's agency example says you can use the approval and brand-check settings to keep each client's creative on-brand without reviewing every single variant by hand.[1] That sentence is the destination, not the first week. It assumes you already know which settings you trust. You earn that by leaving the default on and watching a run.

The direct-to-consumer example is a cleaner picture of the loop, and it is still their example, not a case I ran. A brand on always-on Meta prospecting goes from a handful of hero ads to dozens of hook-and-format variants. The system watches which combinations earn spend and generates the next batch around what is working, without a designer touching every version.[1] Read "without a designer touching every version" as the production claim. Read "approved variants," in the opening paragraph, as the control. Both lines are on the page. The second one goes in your brief.

If you are expanding markets, the page says a hook that wins in one place will not always win in another, and that the system will run the same attribute-level read on each market separately.[1] That is another reason to keep the queue. A localization can be on-brand in the source language and still be the wrong offer in the new market. The brand check, as described here, does not know your market. The person on the queue might.

The one-pager

You do not need the pilot to practice the brief. Write it as a one-pager and keep it next to the ad account. If a teammate later gets into the pilot, this is what they connect.

• Product: Runway Ads, announced 30 September 2026. Pilot with select enterprise partners. Not a public switch.[1]
• Connect: one ad account, one brand kit.
• Kit: guidelines, a chosen set of past ads, product imagery. Add a product feed only if it is already clean enough to generate from.[1]
• Channels named on the page: Meta, Google, TikTok.[1]
• Before you generate: localization rules, including terms that stay untranslated, applied to on-screen text and product screenshots as well as the voiceover.[1]
• Before spend moves: a maximum daily change, a minimum net-new audience share, and a retargeting cap.[1]
• Before anything publishes: the automated brand check, then a person. Human approval stays on.[1]
• Later, and only for one campaign or one variant type: automated publishing, after you have defined what confidence means and watched the queue behave.[1]
• After a yes: read what earned spend. The next round comes from that read, not from a fresh brainstorm that ignores it.[1]
• Do not copy Runway's own ad-account results into your forecast. The page includes them. They are not your baseline.

If a vendor, or a teammate, describes this as "the ads publish themselves," send them back to the default. The page already answered that. Approval is on. You move off it on purpose, for a slice of the program, after the kit and the fences are written.

The creative loop can include a publish step now. Keep the person at the end of it until the queue has earned a shorter line.

Sources

[1] https://runway.com/news/company-news/introducing-runway-ads — Runway, "Introducing Runway Ads," 30 September 2026. The page's visible date is September 30, 2026, and datePublished in the page data is 2026-09-30T18:32:00.000Z. Fetched 1 October 2026.