Marcos Delgado is a high school teacher, not a troll. And in the spring of 2023, he almost destroyed a city council member’s reputation in 48 hours.

A photo landed in his group chat. It showed a local official standing outside what looked like a campaign donor’s private estate. The caption implied corruption. Marcos shared it without thinking. So did eleven other people in the thread. It spread to a neighborhood Facebook group with 4,000 members before anyone stopped to ask a basic question: is this real?

It wasn’t. The image had been generated using Midjourney. The official’s face had been composited onto a stock photo background. By the time the correction circulated, the damage was already done.

Have you ever shared a photo you later second-guessed? Most people have. The uncomfortable truth is that the problem isn’t malice. It’s speed. And the tools built to slow that speed down are sitting unused in browser tabs nobody opens.

Here’s what this actually means for you: three free tools exist right now that can catch manipulated images, AI-generated faces, and deepfake videos before you hit share. I dug into the actual research so you don’t have to. Here’s what I found.

1. Hive Moderation: The AI Detector That Thinks Like a Social Feed

Think of it this way: Hive is like a spell-checker, but for reality. You paste in an image URL or upload a file, and within seconds it returns a confidence score telling you how likely it is that the image was generated by an AI model rather than captured by a camera.

Hive was originally built for enterprise content moderation. The public-facing demo at thehive.ai still works, and it’s free. It’s been stress-tested against outputs from Midjourney, DALL-E, and Stable Diffusion, and in a 2023 independent benchmark by AI forensics researcher Henry Ajder, Hive outperformed six other detection tools on synthetic face identification, hitting above 90% accuracy on uncompressed AI-generated images.

Does the face in this image look just slightly too smooth? That’s often your first instinct, and Hive will confirm or deny it in less time than it takes to type a comment.

Where it earns its place in your toolkit is social media content. Screenshots, shared posts, profile photos used in disinformation campaigns. It’s built for exactly the kind of visual content that moves fast through feeds rather than the kind that arrives as a pristine source file.

Pro Tip: Hive performs best on images that haven’t been heavily compressed. If someone screenshots a screenshot of a screenshot, the AI artifacts get degraded along with image quality. When possible, ask for the original file before running the test. If you can’t get the original, run it anyway. A low-confidence result on a degraded image still tells you something.

One honest limitation: Hive isn’t perfect on images of non-Western faces. The training data skews toward the demographics most represented in English-language synthetic media datasets. That’s a real bias worth knowing about, not a reason to skip the tool, but a reason to use it as one input rather than the only input.

2. FotoForensics: The Tool That Sees What Your Eyes Can’t

FotoForensics uses a technique called Error Level Analysis, or ELA. Here’s the plain-English version of what that means: every time a JPEG image is saved, it loses a little quality. Different parts of an image that have been edited or composited in will show different levels of that compression loss compared to the original parts. ELA makes those differences visible as color variation on the image.

The real story behind the headlines on image manipulation: Bellingcat’s open-source investigators used ELA as part of their toolkit to analyze photographs circulating after the 2014 Malaysia Airlines MH17 shootdown. One specific case involved a photograph purportedly showing a Buk missile launcher in rebel-controlled territory. ELA analysis, combined with geolocation, helped confirm the image hadn’t been digitally altered. The launcher was real. That detail mattered enormously in the subsequent investigation.

You can run the same analysis at fotoforensics.com right now. Upload any JPEG and the tool generates an ELA map. It’s best on faces and on images where something has been added or removed, like a logo, a sign in the background, or a person.

Did You Know: FotoForensics was built and is maintained by Neal Krawetz, a digital forensics researcher who has testified in legal proceedings about image authentication. The tool has been publicly available since 2012. It still works. It’s free. And it’s used by journalists at organizations including Reuters.

The learning curve is real. ELA results are not a simple yes or no. They require some practice reading the output before you’ll feel confident interpreting it. Krawetz maintains a tutorial section on the site that explains what patterns to look for. Spend twenty minutes with it once, and you’ll be able to read results independently.

Ask yourself why media literacy campaigns don’t lead with this tool. A free resource used by professional investigators, available to anyone, barely mentioned in mainstream coverage about misinformation. Convenient, right?

3. Microsoft Video Authenticator: Deepfakes Meet Their Match

Still images are one problem. Video is a larger one. Would you recognize a deepfake of someone you know?

Microsoft Video Authenticator was developed and released in 2020 as part of Microsoft’s AI for Health initiative and later expanded under their Responsible AI work. It analyzes video files frame by frame, looking for the blending artifacts that deepfake generation models leave behind, particularly around facial boundaries, eye movement, and lighting inconsistencies that human perception tends to miss.

In a 2020 evaluation published alongside the tool’s release, Microsoft reported detection accuracy above 96% on a test set drawn from the FaceForensics++ benchmark dataset, one of the most targeted deepfake detection datasets in the field. That number has degraded somewhat as generation models have improved. Researchers at the University of Southern California’s Information Sciences Institute found in 2022 that detection accuracy across tools including Video Authenticator dropped to the 70 to 80 percent range against newer GAN-based deepfakes. Still useful. Not infallible.

Think of it this way: a smoke detector that’s right 75% of the time is still worth having in your kitchen. The goal isn’t certainty. It’s adding friction to the share reflex.

Video Authenticator is best suited for direct video files, things like press releases, official statements, or video evidence someone sends you directly. It’s remarkably precise when the source file is high resolution and uncompressed. Heavily screen-recorded or recompressed clips reduce its effectiveness, which is why pairing it with Hive for still frames from suspicious video is smarter than relying on either tool alone.

Warning: Microsoft Video Authenticator performs best on high-resolution source files. If someone sends you a heavily compressed or screen-recorded clip, run individual still frames through Hive first before drawing conclusions from Video Authenticator’s output. A clean result on a degraded file isn’t a clean bill of health.

Reality Check: A 2024 Reuters Institute Digital News Report found that 62% of people surveyed across 47 countries said they were not confident in their ability to identify what is real or fake online. You’re probably in that group. So is Marcos. So was I, before I started using these tools seriously.

The access point is through Microsoft Azure’s Cognitive Services demo portal. It requires a free Microsoft account rather than a simple URL upload, which is a higher barrier than Hive or FotoForensics. That friction keeps most casual users from discovering it, which is exactly the kind of design choice worth noticing rather than excusing.

Your Next 3 Steps

Step 1: Open the right tool for what you’re looking at right now. If you’re holding a suspicious image from social media, open Hive at thehive.ai first. It’s the fastest, requires no account, and it’s built for exactly the kind of compressed social content you’re likely looking at. If someone sent you a photograph claiming to document a real event, open FotoForensics at fotoforensics.com. If you received a video file directly, go to Microsoft Azure’s Video Authenticator. Matching the tool to the content type is what separates a useful check from a wasted one.

Step 2: Build the habit before you need it. Open all three URLs right now and bookmark them in a single browser folder. Label the folder “Fact-Check Kit.” Do this on your phone, not just your laptop, because that’s where you’re actually sharing things. The twenty seconds it takes to bookmark them now is the only thing standing between you and the kind of situation Marcos found himself in.

Step 3: Test the tools on something real before the next fake arrives. Go find the last suspicious image or video that landed in your messages. Run it through whichever tool matches the content type. Screenshot the result. Send it back to whoever sent you the original. That one action builds a checking habit rather than just an awareness of one, and it signals to your network that verification is a normal thing you do.