Trend report · gnews_detection · 2026-06-16
Fact Check: SRK promotes AI-powered money-making scheme? No, video is a deepfake - NewsMeter
A widely shared video claiming Shah Rukh Khan promotes an AI-powered money-making scheme was confirmed as a deepfake—and it's exactly why platforms are getting aggressive about scanning uploads for AI-generated content. The problem: that same detection net catches legitimate creators using AI editing tools.
What actually flags your file
When you upload a video or image in 2026, platforms don't just look at what's visible. They scan the invisible metadata layer that travels with your file. Here's what's actually being checked.
C2PA / Content Credentials is the cryptographic "made by AI" manifest. It's stored as JUMBF atoms and contains a manifest that explicitly lists the tool that generated or modified the content. If you exported from Sora, Runway, Midjourney, or almost any generative AI tool, your file carries a C2PA manifest that says so. Platforms read this with ExifTool or similar forensic tools before your content ever appears publicly.
XMP metadata includes flags like DigitalSourceType: trainedAlgorithmicMedia—a direct signal that the pixel data came from an AI model rather than a sensor. Generative AI tools write these tags automatically. A raw AI export can carry 144 metadata tags; a phone capture typically has around 94, and they're different tags entirely.
Encoder fingerprints are baked into the bitstream itself. Video encoders like Lavc (FFmpeg's libavcodec), x264, or x265 leave specific SEI (Supplemental Enhancement Information) markers and quantization patterns. AI-generated video exports from tools like Runway or Pika have recognizable encoder signatures that differ from phone capture. Missing encoder fingerprints—or the wrong ones—trigger detection.
GPS, timestamp, and device identity are the absence signals. A phone capture includes Make, Model, Software version, GPS coordinates, and a capture timestamp in the EXIF data. A file generated by Stable Diffusion, DALL-E, Sora, or Flux has none of this. Platforms treat missing location and device data as a strong signal of AI generation.
Perceptual hashes (pHash, aHash) compare your content against known AI-generated baselines. These detect patterns in the pixel generation process itself—not perfect, but improving rapidly as training datasets grow.
How Calabi handles it
Calabi runs a three-stage pipeline that strips the detection signals and replaces them with authentic phone-capture identity. This is not editing or inpainting—your content looks exactly the same. It works on the invisible metadata and bitstream layer that platforms actually scan.
Stage 1 — Strip: Calabi removes all C2PA / JUMBF manifests, zeroing out the C2PA atoms and references. It strips XMP flags including DigitalSourceType: trainedAlgorithmicMedia and any generator/tool namespace tags. It clears Lavc, x264, or x265 SEI markers from video bitstreams. The result: a file with no AI-generation signals remaining.
Stage 2 — Inject: Calabi writes authentic phone-capture identity into the file metadata. You choose a device profile—iPhone 15 Pro, Pixel 8 Pro, Galaxy S24 Ultra—and the file gets Make, Model, Software version, GPS coordinates, and capture timestamp matching that device. The encoder fingerprint is set to match the phone's native encoder. A file that started with zero device identity now reads as a genuine phone recording.
Stage 3 — Verify: Before download, Calabi returns a forensic proof card showing exactly what was stripped and what was injected. This uses the same ExifTool scan platforms use—so you see what moderators and detection systems will see. Verified reductions: 18 JUMBF atoms to 0, 16 C2PA references to 0, the trainedAlgorithmicMedia flag removed, 144 metadata tags reduced to approximately 94 neutral structural tags.
Step-by-step
- Upload your AI-generated video or image file to Calabi.
- Automatic pipeline runs—strip, inject, verify all happen without manual editing or configuration.
- Review the forensic proof card showing exactly what was removed and what phone identity was injected.
- Download the cleaned file with verified neutral metadata and authentic device identity.
FAQ
- Does this work on files that already have visible watermarks?
Calabi removes invisible detection signals and metadata—not visible logos or overlay marks. If a visible watermark is present, cropping it off is the standard approach. Calabi handles the metadata layer that survives cropping and would still get you flagged even after the visible mark is gone. - Can this guarantee I won't get flagged?
No tool can guarantee that. Results vary by platform, source model, and detection method evolution. Calabi removes the structured metadata and encoder signals that are the primary triggers for automated moderation on Instagram, TikTok, YouTube, and Reddit in 2026.
The SRK deepfake is a reminder of why platforms are scanning uploads so aggressively—and why legitimate creators need a way to post AI-edited content without getting caught in that same net. The detection infrastructure is real, it's active, and it's looking at your file's metadata before it ever looks at your pixels.
→ Try Calabi free at calabilabs.com — 10 cleans, no card.