🖼️ Deepfakes & Digital Verification

Curated links and references for spotting AI-generated media and manipulated visuals, and verifying authenticity.

Use this page when a video, image, or audio clip feels “off,” or when a sensational claim spreads with eye-catching media.

✅ Deepfake Fact-Checking

AFP Fact Check — Global fact-checking that frequently debunks manipulated/AI-generated media. ✔IFCN-Verified
🌐 https://factcheck.afp.com ↗️

Reuters Fact Check — Verifies viral visuals and investigates manipulated or AI-assisted content. ✔IFCN-Verified
🌐 https://www.reuters.com/fact-check ↗️

AP Fact Check — Checks claims with a strong focus on political and viral visuals. ✔IFCN-Verified
🌐 https://apnews.com/hub/ap-fact-check ↗️

🧰 Digital Verification Tools & Techniques

InVID & WeVerify — Browser toolkit for video keyframes, reverse-image lookups, and metadata checks.
🌐 https://www.invid-project.eu ↗️

TinEye Reverse Image Search — Finds earlier/alternate versions of images to trace first appearance.
🌐 https://tineye.com ↗️

Google Images (Search by Image) — Reverse-image search via camera/upload to find sources and context.
🌐 https://images.google.com ↗️

Forensically (29a.ch) — Visual forensics suite (clone detection, error levels, noise analysis).
🌐 https://29a.ch/photo-forensics ↗️

Wayback Machine — See prior versions of pages/posts; verify if media/claims existed before an event.
🌐 https://web.archive.org ↗️

📌 Sam’s Tips

Extract keyframes before searching. One thumbnail won’t surface the truth—search multiple clear frames.

Chase the first appearance. If the image predates the “news,” it’s repurposed—or worse.

Edges and physics betray fakes. Watch hair/earlines, reflections, jewelry, shadows, and hand/finger geometry.

Audio tells on close listen. Robotic cadence, breath/noise floor mismatches, and sudden accent/intonation shifts are red flags.

Metadata helps, but chain-of-custody matters more. Screenshots strip EXIF*—so lean on cross-checks and provenance.

*Exchangeable Image File Format. Metadata that many image files (JPEG, TIFF, HEIC, some WEBP/RAW) store inside the file, including useful clues about when/where and how a photo was made.

🖼️ Quick example

A video spreads online showing a well-known public figure apparently saying: “The country is collapsing — take your money out now.”

Stop — alarming or sensational statements are common deepfake hooks.
Investigate — the video appears only on social media accounts with no links to reputable outlets.
Find better coverage — AP, Reuters, and AFP show no reporting on the quote, and no official transcript includes it.
Trace — extracting keyframes and reverse-searching them reveals the footage is from an unrelated interview recorded years earlier.

Conclusion

The claim fails verification. The footage originates from a real interview, but the audio has been altered and the context misrepresented.

Why

• Sensational statements are frequently used in manipulated clips to trigger fear.
• Independent fact-checkers and reputable outlets did not report this quote.
• Keyframe searches reveal the original, unedited source, confirming the mismatch.
• Deepfake audio is often inserted into real footage to make the clip more believable.

Bottom line:
If a dramatic video appears only on social platforms and can’t be matched to a reputable original source, treat it as unverified until proven otherwise.