Calabi Labs · Guide · 2026-06-14
```html
If you searched "background people remover," you're probably looking for a tool that can detect a random stranger in your vacation photo and vanish them like they were never there. That's a visual editing task — inpainting, content-aware fill, object removal. Tools like Photoshop's generative fill, Snapseed, or dedicated object removers handle that. Calabi does not do visual editing. It works on an entirely different layer.
Calabi is a forensic cleaning tool for AI-generated content. If you're exporting a video from Sora, Runway, or Midjourney and getting flagged, suppressed, or labeled as "AI-generated" by Instagram, TikTok, or YouTube — that's a metadata and signal problem, not a pixel problem. Calabi strips the invisible detection signals and injects authentic phone-capture identity so platforms read your file as a normal phone recording. If you need to remove a visible person from a photo, look for a photo editor. If you need AI-generated content to stop getting flagged, Calabi is built for that.
Platforms don't flag content because of what it looks like — they flag it based on invisible metadata and encoding signals embedded in the file itself. When you export from an AI generator, the file carries a forensic trail that automated systems are specifically trained to detect.
The most damning signal is C2PA / Content Credentials — a cryptographic manifest stored as JUMBF atoms that explicitly declares the content was created by a specific AI model. This is the "Made with [Tool]" badge that Adobe, Microsoft, and the Coalition for Content Provenance and Authenticity are pushing industry-wide. If your Sora export carries C2PA atoms referencing generative AI, platforms read that manifest before they even render a single frame.
Beyond C2PA, there's the DigitalSourceType: trainedAlgorithmicMedia XMP tag — a direct flag that the file came from a model trained on scraped data. Encoder fingerprints are another giveaway. AI video exporters use libraries like Lavc (FFmpeg's libavcodec) or x264 with specific SEI (Supplemental Enhancement Information) headers that leave a statistical fingerprint distinct from phone hardware encoders. A phone records with Apple A-series or Qualcomm Snapdragon silicon. An AI export records with software encoding that screams synthetic.
Missing GPS coordinates, a capture timestamp from six months ago, and a device make/model of "Unknown" round out the detection profile. A real iPhone 16 Pro capture has GPS, a recent timestamp, and an encoder profile that matches Apple's hardware. An AI export has none of that — and platforms know what to look for.
Here's the hard part: even if you visually edit your AI content — crop out a watermark, remove a person, run it through a filter — the metadata and encoding signals survive. The C2PA manifest is embedded in the file, not the pixels. Screenshots strip some metadata but leave the encoder fingerprints. Re-uploading through a different platform still carries the Lavc or x264 bitstream signature. The detection layer is not visual — it's structural.
Photo editing tools remove things from images. Calabi removes the detection signals that make platforms flag your file in the first place. They're solving different problems.
Calabi runs your file through a three-stage pipeline that targets every signal platforms scan for:
Can Calabi remove a visible watermark or logo from my image?
No. Calabi does not edit pixels, remove objects, or perform inpainting. If you need to crop out a visible watermark, use a photo editor like Photoshop, GIMP, or an online object remover. Calabi removes the invisible detection layer — the C2PA manifests, XMP flags, and encoder fingerprints — that survive cropping and survive most re-encodes. That's the layer platforms actually scan for.
Does cleaning my file guarantee a platform won't flag it?
No tool can guarantee zero flags on any platform. What Calabi does is eliminate the structural signals that automated systems specifically look for. Results vary by platform, source model, and how the file has been processed. Calabi gives you a verified clean file with authentic phone identity — that's the strongest possible foundation for passing platform scans.
What's the difference between Calabi and a screenshot or screen recording?
A screenshot strips some metadata but often preserves the encoder fingerprint of the capture device (your computer screen recorder) and adds its own artifacts. A screen recording of AI content still carries Lavc or x264 fingerprints and lacks GPS/timestamp/phone identity. Calabi produces a file with verified iPhone or Pixel hardware encoder identity, accurate GPS and timestamp, and zero AI manifest references — a clean forensic profile, not just metadata reduction.
Try Calabi free at calabilabs.com — 10 cleans, no card.