Runway video detector

From creative tool to everyday footage

Runway has been building AI video tools since before the current boom — it co-created the original Stable Diffusion research and released Gen-1 and Gen-2 in 2023, followed by Gen-3 Alpha and Gen-4 with much better consistency of characters and scenes across shots. It is widely used in advertising, music videos and independent film, and its output increasingly appears in mainstream media as b-roll, title sequences and visual effects.

That professional use shapes how its clips look: deliberate camera moves, cinematic colour grading, shallow depth of field and consistent characters across several shots. Those same qualities make stock-like Runway footage easy to slip into a video essay or news-style clip without anyone noticing.

How to check a Runway clip

Start with the original export if you have it: GPTTrace reads Content Credentials from MP4 files and shows the signer and the recorded actions. Without credentials, it samples eight frames and runs the image classifier and spectral checks; the frame strip shows the score of each, which helps when a video mixes generated shots with real footage, as many edits do.

When you can only see the clip on a platform, look for continuity problems between cuts, text in the scene that never quite resolves into words, and physical interactions — pouring, catching, opening doors — that look slightly wrong. Credits are also worth reading: professional work made with Runway often says so.

Disclosure for creators

If you make work with Runway, keeping Content Credentials on your exports and noting AI use in the description protects you as platforms tighten their labelling rules. Several platforms now ask uploaders to disclose realistic AI-generated video, and some apply labels automatically when they find credentials. Credentials also give you a dated record of your own work if someone else reposts it.

Shots mixed with real footage

Professional edits rarely use generated footage alone. A typical music video or ad cuts between real performance footage, stock video and a few generated shots for scenes that would be expensive or impossible to film. Frame sampling helps here: if most frames score low and one or two score high, scrub to those timestamps and look closely. A sudden change in grain, colour response or lens character between cuts is a further hint that a shot came from somewhere else.

For news and documentary contexts, the bar is different. Generated footage presented as a record of real events is misleading even when it is only a few seconds of a longer piece. If you work in those fields, ask for the source and the rights information of every shot, and check frames from any shot without a clear origin.

Generated shots are also common in YouTube video essays as illustrative b-roll.