Calabi Labs · Guide · 2026-06-14
Remove logo from video
Here's the page:
```html
How to Remove a Logo from a Video — What Actually Works
You can remove a visible logo from a video using an inpainting tool like Morph Studio, Wink, or Airbrush — those tools paint over the pixel region where the logo sits. Calabi does something different: it strips the invisible forensic signals that get you flagged by Instagram, TikTok, YouTube, and Reddit even after you've already cropped or erased the visible mark. If you want to understand both layers — and why the invisible one is what actually gets you banned — keep reading.
What Actually Gets Your Video Flagged
When you upload a video, platforms run an automated scan before it ever appears publicly. That scan isn't just looking at the pixels — it's reading the metadata and forensic signals embedded in the file itself. Three things trigger automatic detection:
- C2PA / Content Credentials — This is a cryptographic manifest embedded in the file as JUMBF atoms. It records which AI model generated the content, when, and with what tool. Platforms like Instagram and YouTube read this manifest automatically and use it to label or throttle AI-generated uploads. A Sora export, a Runway clip, or a Pika generation carries this manifest by default.
- XMP AI flags — Specifically the
DigitalSourceType: trainedAlgorithmicMediatag. This single XMP field tells any forensic scanner "this came from an AI model trained on scraped data." It's part of the IPTC XMP schema and survives re-encoding, recompression, and format conversion unless explicitly stripped. - Encoder fingerprints — AI video exports from models like Sora, Runway, and Kling carry recognizable encoder signatures in the bitstream — Lavc (FFmpeg's libavcodec), x264 SEI messages, and similar tool印记. These fingerprints persist even when you re-encode the video, because they're written into the compressed bitstream structure, not just the metadata headers.
On top of those three, platforms also check for absence signals: no GPS coordinates, no capture timestamp, no real device metadata. A file that has AI generation markers but no phone-capture identity looks machine-made to automated classifiers.
Why the Obvious Fixes Fail
If you've already tried cropping the logo out, screenshotting the video, or re-uploading after re-encoding, you may have noticed the platform still flags it. That's because those methods only address the visible layer. Here's what actually happens in each case:
- Cropping — Removes the visible logo pixels from the frame. The file metadata, C2PA manifest, and encoder fingerprints are completely untouched. Platforms scan the file, not just the frame.
- Screenshotting — Re-renders the video as an image sequence, which drops some metadata. But the XMP AI flags and C2PA atoms often survive in the new file, and the encoder fingerprint of whatever tool you used to screenshot it gets recorded instead.
- Re-encoding / recompressing — This removes some metadata headers but almost never touches the C2PA manifest or XMP tags embedded in the file's core structure. The trainedAlgorithmicMedia flag has been found in files re-encoded multiple times.
In short: removing the visible logo and removing the detectable AI signals are two separate problems. Most tools solve the first one. Calabi solves the second one.
How to Clean the Invisible Detection Layer
Calabi works on the file-level forensic signals — the layer that survives cropping, screenshotting, and re-encoding. Here's what the pipeline does:
- Upload your video — Drag and drop or select the file. No manual settings, no region selection, no pixel editing.
- Automatic strip + inject — Calabi's pipeline runs in one pass: it strips every detectable C2PA/JUMBF atom (reducing 18 C2PA references to 0), removes the DigitalSourceType: trainedAlgorithmicMedia XMP tag, strips encoder fingerprints like Lavc and x264 SEI headers, and eliminates tool-specific metadata. Then it injects authentic phone-capture identity: a real device profile (iPhone 15 Pro, Pixel 8 Pro, or Galaxy S24 Ultra), a capture timestamp, GPS coordinates, and a genuine phone encoder name.
- Review the forensic proof card — Before downloading, you see the same ExifTool scan platforms use. It shows exactly what was stripped (the AI signals) and what was injected (the phone identity). This is the record that proves the file reads as a normal phone recording.
- Download the cleaned file — The file is ready for upload on any platform without triggering the automated AI-detection scan.
FAQ
Can Calabi remove a visible logo or watermark from my video?
No. Calabi does not paint over, erase, or reconstruct any pixels in your video frame. If you have a visible logo, Sora's sparkle mark, or a platform watermark you need gone, use a dedicated inpainting tool like Morph Studio, Wink, or Airbrush first. Then run the cleaned file through Calabi to strip the invisible detection signals that still get you flagged.
Why does my video still get flagged even after I cropped the logo?
Because cropping removes the visible logo but leaves the file's forensic metadata completely intact. The C2PA manifest, XMP AI flags, and encoder fingerprints are embedded in the file structure, not the frame composition. Platforms scan the file, not the pixels — so a cropped Sora export still carries every detectable AI signal.
What does Calabi's forensic proof card actually show?
It shows a before-and-after ExifTool readout. You can see the 18 JUMBF/C2PA atoms that were reduced to 0, the DigitalSourceType: trainedAlgorithmicMedia flag that was removed, the Lavc and x264 SEI encoder fingerprints that were stripped, and the injected device profile — Make, Model, Software, GPS, and timestamp — that makes the file read as a phone recording. This is the same forensic scan that newsrooms and platform trust-and-safety teams use.
Try Calabi free at calabilabs.com — 10 cleans, no card.
```