Since Sora 2 landed at the end of September 2025, near-photoreal AI clips have gone from novelty to background noise in most feeds. The good news is that OpenAI, unlike some rivals, ships every video with two provenance markers on purpose: a watermark you can see and a signature you can't. The bad news is that both can disappear by the time a clip reaches you β sometimes because someone removed them, more often just because sharing it stripped them out.
So the honest answer to "is this a Sora video?" is layered. When the markers are there, they're close to conclusive. When they're gone, you're back to reading the footage and checking the story. Here's the order I'd work through it.
The 6-step check
Look for the moving watermark first
Every clip downloaded from sora.com or the Sora app carries a visible watermark: the small white Sora cloud icon that bounces around the edges of the frame, changing position every second or two. That drift is deliberate β a static logo in one corner is trivial to crop out, so OpenAI made it move. Play the clip full-screen and watch the borders for a few seconds. If you catch the icon skating along the edge, you're almost certainly looking at a Sora export. Reposted clips are usually lower resolution, so pause on a few frames if it's hard to spot in motion.
Treat a missing watermark as inconclusive, not proof
This is the trap. No watermark does not mean the video is real. An entire cottage industry of "Sora watermark remover" sites and apps sprang up within weeks of launch, and the moving logo can be cropped, blurred, painted over, or covered with an overlay by anyone willing to spend a minute on it. A video with no visible mark could be a genuine camera recording β or a Sora clip that's been cleaned up. So the watermark is a strong positive signal when it's present and tells you almost nothing when it's absent. When it's gone, move to the metadata.
Inspect the C2PA Content Credentials
Alongside the visible mark, OpenAI embeds C2PA metadata β an industry-standard provenance signature (the same "Content Credentials" standard Adobe, Google, and camera makers back) β in the file, flagging it as AI-generated by Sora. Unlike a cropped logo, this signature is designed to survive inside the file itself. To read it, drop the video into the Content Authenticity Initiative's free Verify tool at contentcredentials.org/verify. If the file still carries its credentials, you'll see the issuer and a note that it was made with an AI tool. Present credentials are close to a confirmation; the next step covers why absent ones prove nothing.
Remember that metadata gets stripped in normal sharing
C2PA data is durable inside the original download, but most social platforms and messaging apps re-encode uploads and throw the metadata away in the process. A screen recording of a Sora clip, a re-upload to a video app, or a screenshot of a frame will all typically arrive with no credentials at all β not because the content is authentic, but because the pipeline it traveled through deleted the evidence. So by the time a viral clip reaches your feed, both the watermark and the metadata may be long gone. Absence of provenance is the normal state for anything that's been reshared, not a green light.
Read the clip for Sora's own tells
When both markers are missing, fall back on what the footage shows. Current AI video still fumbles the boring physics: liquids that pour wrong or refill themselves, hands and fingers that morph between frames, text on signs that scrambles into gibberish, reflections that don't match the scene, and objects that pop in or vanish at the edges. Sora clips also tend to run only a handful of seconds, sit in a clean 9:16 or 16:9 aspect ratio, and carry an oddly smooth, cinematic sheen with no camera shake. None of these is proof on its own β real footage can look weird too β but a cluster of them on a suspiciously perfect clip is a reason to keep checking.
Verify the claim, not just the pixels
Most of the harm from a synthetic clip comes from the story attached to it β a supposed event, a quote put in someone's mouth, a "leaked" moment. Even if you can't nail down whether a specific file is Sora, you can check the claim. Has any credible outlet reported the event? Does a reverse image search on a keyframe turn up the real, original footage or an earlier date? Are fact-checkers already writing about it? That last one is fast: FAXTR searches 100+ fact-checking organizations in one query, so you can see whether the clip circulating in your feed already has a published verdict before you share it.
Signals worth training your eye on
You won't run a full metadata check on every clip that crosses your feed. These are the tells worth noticing at a glance β any one of them means the video is worth a closer look before you believe or forward it.
- βA small white cloud icon drifting around the edges of the frame β the giveaway that a clip was exported from Sora.
- βA short, ~10-second clip in a perfectly clean vertical or widescreen ratio with no camera shake or ambient noise.
- βWarping hands, morphing objects, or on-screen text that dissolves into nonsense when you pause.
- βA dramatic "breaking" moment that no news outlet has reported and that only exists as this one video.
- βA blurred, cropped, or oddly-overlaid corner where a watermark could have been removed.
- βContent Credentials that name an AI generator when you run the file through a C2PA checker.
Why "no watermark" keeps fooling people
It's natural to reason that if a clip had the Sora logo it'd be obvious, so a clean-looking video must be a real recording. That inference is exactly backwards. The moving watermark is easy to crop and there are apps built to erase it, while the invisible C2PA signature is quietly deleted by ordinary re-uploads and screen recordings. The result is that the most deceptive synthetic clips β the ones stripped, cropped, and reshared until they look native to your feed β are precisely the ones with no markers left. Absence of provenance is a prompt to check harder, never a stamp of authenticity.
What this guide is not
This isn't a claim that Sora footage is inherently deceptive β most of it is people making jokes and short films, clearly labeled. And these checks won't catch every model; a clip from a different generator won't carry Sora's watermark at all. The point is a repeatable habit for the moment a "real" video makes a big claim: read the markers if they're there, read the footage if they're not, and check the story either way before it travels any further on your say-so.
Check the claim in seconds
A viral clip making a big claim? FAXTR searches 100+ fact-checking organizations in one query β free, no login required.