Runway Risks, Limits and Governance Rules to Set Before You Scale

Runway, built by the New York company Runway AI, offers a family of video generation models alongside editing tools such as background removal, motion tracking, inpainting and frame extension. A marketer can upload a product photo, describe camera movement, and get a few seconds of animated footage. Other tools in the suite remove objects from a shot or extend a clip beyond its original frame.

The practical result is that storyboards, social cutdowns and concept videos no longer need a full shoot. That speed is the reason governance becomes urgent. When a clip takes three minutes instead of three days, review steps that used to happen naturally on set disappear unless someone puts them back on purpose.

Every generative video tool has limits, and planning around them saves budget. With Runway, the ones that affect marketing work most are well known among users:

• Clip length is short per generation, so longer pieces are stitched from several outputs and continuity can drift between them.
• Hands, text on packaging and fine logos can distort, which makes product close-ups risky without manual correction.
• Faces and characters may change subtly between shots unless you use reference images and consistency features.
• Generation consumes credits, and iteration is where most credits go, not the final render.
• Physics is plausible rather than accurate; liquids, fabric and reflections often need a second pass or a real plate.

None of these make the tool unusable. They mean Runway works best for mood pieces, transitions, backgrounds and concept tests, and less well as a replacement for a hero product shot that legal will scrutinise frame by frame.

Start with the risks that create liability, not the ones that create bad clips. A bad clip gets deleted. A clip that uses someone’s likeness without permission can become a dispute.

Rights and inputs

Anything you upload becomes part of the generation. If a designer feeds in a stock image licensed only for print, or a competitor’s ad as a style reference, the output inherits that problem. Your rule should be simple: only owned assets, properly licensed stock, or material created inside the tool go in.

Likeness and consent

Generating a realistic person who resembles a real employee, influencer or public figure is the fastest way into trouble. Write down that no real person’s face or voice is used without a signed release that explicitly covers synthetic media. Keep that release with the project files.

Disclosure

Platforms are adding labels for AI-generated or altered media, and several ad networks ask advertisers to declare it. Decide internally when you label, rather than relying on each editor’s judgement. A reasonable default is to disclose whenever generated footage shows people or realistic events.

Brand accuracy

Generated products can gain buttons, lose stitching or change colour. A claim-bearing visual that shows a feature the product does not have is a consumer protection issue, not just a design flaw.

The fix is a short, mandatory gate rather than a long committee. In a typical team, a workable flow looks like this:

• The brief states the intended use: social organic, paid ad, website or internal concept only.
• The creator logs prompts, reference files and seeds in a shared sheet next to the output file name.
• A second person checks the clip against a five-point list: rights of inputs, people, product accuracy, text legibility, disclosure need.
• Paid placements get an extra sign-off from whoever owns brand or legal risk.
• Approved files move to a separate folder; nothing is published from the working folder.

The log is the part teams skip and later regret. When a platform or a client asks how a clip was made, a prompt record answers in seconds.

Keep the policy to one page so people actually read it. The rules that earn their place are these:

• Approved uses, listed by channel, plus a short list of banned uses such as fake testimonials or simulated news footage.
• Input rules: owned or licensed only, no client confidential material in prompts unless the contract allows it.
• People rules: no real likeness without written consent, no minors, no generated “customers”.
• Labelling rules matched to each platform you advertise on.
• Retention: keep prompts and source files for as long as you keep the campaign assets.
• Account control: company-owned workspace, named seats, no personal accounts for client work.

Account control sounds administrative, but it is what stops a freelancer leaving with every project file and generation history.

Policies drift. A deadline arrives, someone skips the second check, and the shortcut becomes habit. Audit a sample of published clips every quarter against the log and you will catch drift early.

Model updates also change behaviour. A prompt that produced a clean result last month may give something different after an update, so treat saved prompts as starting points rather than guarantees. Finally, terms of service change. Read the current Runway terms on commercial use and data handling yourself, at the time you sign up, rather than relying on a summary written months earlier.