Trend report · gnews_detection · 2026-06-19
VerifyLabs.AI Brings Deepfake Detection to Android After a recent IOS release - IT Security Guru
Deepfake Detection Is Now on Your Phone — And That's a Problem for AI Creators
VerifyLabs.AI just launched its deepfake detection tool on Android, following an iOS release. That's one more scanner in the pipeline platforms use to flag AI-generated content. If you're posting AI videos or images on Instagram, TikTok, YouTube, or Reddit, detection tools like these are increasingly scanning uploads automatically — often within seconds of upload. The question isn't whether your content looks like a deepfake. It's whether your file signals "AI-generated" to the automated systems doing the initial sweep.
What Actually Flags Your File
Platforms in 2026 don't rely on a single check. They run a layered scan across multiple forensic signals embedded in your file itself — not just what's visible on screen.
The first layer is C2PA / Content Credentials — the cryptographic manifest that stores "made by AI" provenance data in JUMBF atoms. When you export from Sora, Runway, Midjourney, or Leonardo AI, these tools write a C2PA manifest into the file metadata. It includes fields like digitalSourceType=trainedAlgorithmicMedia and a list of Content Credentials references. A single AI export can contain 18+ JUMBF atoms and 144 metadata tags. Instagram and TikTok both scan for this. If your file still carries it, you're flagged before a human moderator ever sees your post.
The second layer is XMP AI metadata — fields like DigitalSourceType, Generator, SoftwareAgent, and creator tool tags embedded in the XMP packet. Even after you strip C2PA, these XMP flags can remain. YouTube's content ID system and Reddit's automated moderation both query XMP fields as part of their initial triage.
The third layer is encoder fingerprints. Video files carry codec signatures — fields written by the encoding software. Lavc (libavcodec), x264, x265, and specific SEI (Supplemental Enhancement Information) messages in H.264/H.265 streams are recognized as "generated" rather than "captured." A phone recording uses a real device encoder like AppleVEDevice or qcom.Encoder. An AI export uses Lavc or a cloud encoder. That difference is fingerprintable.
The fourth signal is missing phone identity. A genuine phone capture includes Make, Model, Software version, GPS coordinates, and capture timestamp in EXIF. AI exports typically lack all of this. Platforms treat absent GPS and device identity as a red flag — especially on Instagram Reels and TikTok, where native phone captures are the expected baseline.
How Calabi Handles It
Calabi runs a three-stage pipeline that strips every detection signal and injects authentic phone-capture identity in its place. This isn't editing the visual content — it's rewriting the file's forensic identity at the metadata and bitstream level.
Stage 1 — Strip: Calabi removes all C2PA / Content Credentials JUMBF atoms (18 atoms down to 0), strips C2PA references (16 references to 0), deletes the trainedAlgorithmicMedia flag, and wipes XMP AI fields. It also targets encoder fingerprints — removing Lavc/x264 SEI markers and other codec signatures that mark a file as machine-generated rather than device-captured.
Stage 2 — Inject: Calabi writes fresh phone identity into the file. You can choose a device profile — iPhone 15 Pro, Pixel 8 Pro, Galaxy S24 Ultra — and the file receives matching Make, Model, Software version, GPS coordinates, and capture timestamp. The encoder fingerprint switches from Lavc to a real-phone codec identifier. The result is a file that looks, to automated scanners, exactly like a native phone recording.
Stage 3 — Verify: Before you download, Calabi generates a forensic proof card — the same ExifTool scan that platforms use. You see exactly what was stripped (all C2PA atoms, AI flags, encoder fingerprints) and what was injected (device profile, GPS, timestamp). This is the same verification tool newsrooms and forensic analysts use to verify file provenance.
A Note on Visible Watermarks
If your AI export has a visible watermark — a corner logo, a sparkle mark — cropping removes it. Calabi doesn't erase visible marks pixel-by-pixel. What it does is strip the invisible detection layer — the metadata and encoder signatures — that survives cropping and re-encoding. That's the layer that actually gets you flagged on platforms, not the watermark you can see.
How Calabi Works — Step by Step
- Upload your AI-generated video or image. Drag and drop or select the file. No manual settings required.
- Calabi runs the automatic strip + inject pipeline. It removes C2PA/JUMBF atoms, XMP AI flags, and encoder fingerprints. It injects a device profile (iPhone, Pixel, or Galaxy) with matching GPS, timestamp, and codec identity.
- Review the forensic proof card. See exactly what was stripped and what was injected — the same ExifTool output platforms read. Confirm the file now carries 0 C2PA atoms and 0 AI flags.
- Download the cleaned file. Your file is now at the forensic level identical to a native phone recording. Ready to upload on Instagram, TikTok, YouTube, or Reddit.
FAQ
Will this guarantee my post won't get flagged?
No tool can guarantee a platform won't flag you — results vary by platform, source model, and detection method. Calabi removes the metadata and encoder signals that automated scanners check in the initial pass. A re-encode also disrupts some invisible pixel watermarks, though results vary. The metadata and C2PA layer is fully stripped.
Does Calabi change how my image or video looks?
No. Calabi works entirely on invisible file metadata and bitstream signatures. Your visual content stays exactly the same. It rewrites the file's forensic identity without touching the pixels.
What's the difference between stripping metadata and re-encoding?
Re-encoding a video degrades quality and doesn't reliably remove C2PA manifests or encoder fingerprints. Calabi surgically targets the specific forensic signals platforms scan for, preserving your original quality while replacing your file's entire identity with authentic phone-capture signals.
→ Try Calabi free at calabilabs.com — 10 cleans, no card.