Deepfake detection tools
How we wrote this
We have not run our own accuracy tests. This guide summarizes what each company says its product does, plus independent research and reporting, all linked below. Vendor accuracy claims are the vendor's own and are labeled as such. No company paid to be included, and none of the links are affiliate links.
At a glance
Consumer tools are built for individuals. Enterprise tools are sold to companies, platforms and governments. Both means there is a free consumer option and a paid business product.
| Tool | Who it's for | What it does | Access | Worth knowing |
|---|---|---|---|---|
| Hive | Both | Free Chrome extension (images, video, audio, text); paid API for platforms | Extension is free with no login; API is paid | In December 2024 Hive announced a contract with the Defense Innovation Unit (US Department of Defense) |
| Reality Defender | Enterprise | Detection API and platform for banks, call centers and governments | Free developer tier: 50 detections a month, image and audio only; video on paid plans | Built mainly for businesses that need to screen calls and media at scale |
| Sensity | Enterprise | Forensic platform for face swaps, voice and document fraud | Enterprise and government; cloud or on-premises | Produces forensic reports for investigations |
| Intel FakeCatcher | Enterprise | Analyzes subtle "blood flow" color changes in video of faces | Not sold as a consumer app | Intel claims 96% accuracy; a 2023 BBC test found it flagged some real, low-quality videos as fake and it does not analyze audio |
| McAfee Deepfake Detector | Consumer | Flags likely AI-generated audio in videos you watch | Runs on-device on select Copilot+ PCs; English audio | Preinstalled on some HP and other laptops; audio only |
| Content Credentials Verify | Free check | Reads C2PA provenance data on a file | Free, browser-based | Only works if a credential is attached; most files have none |
| Google SynthID (in Gemini) | Free check | Checks for Google's invisible AI watermark | Free in the Gemini app | Only detects content made or edited with Google AI tools |
| TrueMedia.org | Shut down | Was a free nonprofit detector for journalists and the public | Closed January 2025; code released as open source | No longer available as a service |
For individuals
- Hive offers a free Chrome extension that scans images, video, audio and text on the page you're viewing, without signing in. It's the most accessible option for a quick check.
- McAfee Deepfake Detector listens for AI-generated audio in videos playing on your PC. It runs on-device and only on select Copilot+ PCs, so most people can't use it on their current computer.
- Video calls: Zoom says it is launching deepfake risk detection in December 2026. Treat any such alert as a prompt to verify, not a guarantee.
For organizations
Banks, newsrooms, platforms and governments typically use APIs that screen uploads, calls or identity checks at volume. Reality Defender, Sensity and Hive's API are the best-known names; Reality Defender also has a free developer tier for testing image and audio detection. Intel's FakeCatcher took a different approach, looking for blood-flow signals in faces, but a BBC test in 2023 found it could mislabel real, low-resolution video.
When evaluating any vendor, ask for results on your kind of media (compressed phone video, call audio, ID selfies), the false-positive rate as well as the detection rate, and how quickly the model is updated for new generators.
Provenance checkers
Provenance tools don't guess whether something looks fake; they read a signed record of where it came from. Content Credentials Verify reads C2PA data attached by cameras, editing apps and AI generators. Google's SynthID detector in Gemini checks for Google's own watermark. Both are free and useful, but a "nothing found" result is the norm for real and fake files alike. See how to check provenance.
What happened to TrueMedia
TrueMedia.org, a nonprofit that offered free deepfake checks to journalists and the public during the 2024 elections, shut down in January 2025 and released its detection code as open source. If an older article recommends it, that option is gone.
The limits
The Deepfake-Eval-2024 benchmark tested detectors on deepfakes that actually circulated online in 2024. Open-source detectors lost about 50% of their measured performance (AUC) on video, 48% on audio and 45% on images compared with older academic test sets. Commercial models did better, but none reached 90% accuracy, and the researchers found common weak spots: background music, edits to only part of a face, and text overlays. In short, a detector trained on last year's fakes may miss this year's.
How to use them
- Start with the source: reverse search and context checks catch most viral fakes. See our spotting guide.
- Use the best copy you can find; screenshots and re-uploads destroy evidence.
- Try more than one detector and note where they disagree.
- Read a score as a probability, never a verdict. "87% likely AI" is a reason to keep checking.
- For money or safety decisions, verify through a trusted channel no matter what a tool says. Our voice-clone scam guide explains how.
Protecting your identity
Detection is only half the defense. Reducing what scammers can find about you makes impersonation harder:
- Data removal services such as Incogni or Optery send opt-out requests to people-search sites for a fee. You can also do this yourself for free.
- Identity monitoring services such as Aura alert you to credit and account activity after a breach.
- A password manager with two-factor authentication limits the damage if one password is phished.
We have no affiliate relationship with any product named on this page. See our affiliate disclosure.
Frequently asked questions
What is the best free deepfake detector?
There is no single best one. For a quick consumer check, Hive offers a free Chrome extension that scans images, video, audio and text on web pages without a login, and Reality Defender has a free developer tier for image and audio. Use more than one tool and verify the source as well.
How accurate are deepfake detectors?
Vendors publish their own accuracy figures, but independent testing is much less flattering. The Deepfake-Eval-2024 benchmark, built from real fakes circulating in 2024, found that no commercial model it tested reached 90% accuracy and open-source detectors lost roughly half their measured performance compared with older benchmarks.
Can a detector prove that something is real?
No. A "likely real" result only means the tool did not find the patterns it was trained on. New generators, compression, cropping, music and text overlays can all fool detectors. Provenance (Content Credentials) and source tracing give stronger evidence when available.
Has DeepfakeCentral tested these tools?
Not yet. This page summarizes what each company says its product does and what independent reporting and research have found, with links. We have not run our own accuracy tests, and we will say so clearly if and when we do.
Sources
Every factual claim on this page links to one of these. We checked each source on Oct 23, 2026.
- Deepfake-Eval-2024: A Multi-Modal In-the-Wild Benchmark of Deepfakes Circulated in 2024 · arXiv, 2025
- AI-Generated Content Detection · Hive
- Reality Defender · Reality Defender
- Sensity AI · Sensity
- Trusted Media: Real-time FakeCatcher for Deepfake Detection · Intel
- Intel's deepfake detector tested on real and fake videos · BBC, July 2023
- McAfee Deepfake Detector · McAfee
- Verify Content Credentials · Content Authenticity Initiative
- SynthID · Google DeepMind
- TrueMedia.org · TrueMedia