Calabi Labs · Guide · 2026-06-13
Red eye remover
A red eye remover is a photo editing tool that fixes the red/purple reflection in eyes caused by camera flash bouncing off the retina. Traditional red eye removers work by detecting the red pixels in the affected area and replacing them with a natural skin tone or dark pupil color. However, if you're an AI content creator looking to post generated images on Instagram, TikTok, or Reddit, the "red eye" problem you're actually facing is invisible — it's the metadata fingerprints and encoder signals that get your file flagged, not anything you can see in the pixels.
Calabi is not a red eye remover in the traditional sense. It does not edit pixels, adjust colors, or use inpainting tools. What Calabi handles is the invisible layer — the metadata and bitstream signals that platform scanners detect within seconds of upload. If you generated an image with Midjourney, Sora, DALL-E, or any other AI tool and want that file to read as a normal phone recording at the forensic level, Calabi is built for exactly that.
What actually gets flagged
When you upload an AI-generated image, platforms don't just look at what the picture contains. They scan the invisible metadata layer and the file's structural signatures. The most common flags include C2PA / Content Credentials stored as JUMBF atoms — this is a cryptographic manifest that explicitly declares "this image was AI-generated" with a full provenance chain. The XMP tag DigitalSourceType: trainedAlgorithmicMedia is another red flag that gets checked. Encoder fingerprints like Lavc (from video exports) or x264 SEI messages in video bitstreams tell scanners the file was machine-generated rather than captured by a phone. A file missing GPS coordinates, capture timestamp, and phone-specific metadata also signals synthetic origin. Raw AI exports can carry 144+ metadata tags — platforms know exactly what generator made your file and when.
Why cropping or screenshotting doesn't solve it
The instinct is to crop out the problem or take a screenshot to "reset" the file. Screenshotting does strip some metadata, but platform scanners have gotten sophisticated — they also check perceptual hashes, encoder fingerprints, and structural patterns that survive re-encoding. Cropping removes visible artifacts but leaves the metadata layer completely intact. That cropped AI image still carries its original generation timestamp, generator tool tags, and Content Credentials manifest. Re-exporting through Photoshop or another tool might strip some EXIF data, but C2PA signatures embedded in the JUMBF structure survive most consumer editing workflows. The invisible "red eye" — your file's AI fingerprints — persists through all of these workarounds.
How Calabi actually cleans your file
Calabi runs a three-stage pipeline on your upload. First, it strips: all C2PA / Content Credentials JUMBF atoms are removed, the DigitalSourceType: trainedAlgorithmicMedia XMP flag is deleted, generator and encoder tags are stripped, and Lavc/x264 SEI fingerprints are neutralized. Second, it injects: a real phone device profile — iPhone 15 Pro, Pixel 8 Pro, or Galaxy S24 Ultra — with authentic Make, Model, Software version, GPS coordinates, and capture timestamp. The encoder name becomes a real phone encoder instead of "Lavc" or "x264." Third, it verifies: before download, you receive a forensic proof card showing exactly what was stripped and what was injected, verified with ExifTool — the same tool newsrooms and platform compliance teams use.
The result is a file that reads at the forensic level as a normal phone recording. The visible content — your AI-generated image — stays exactly the same. It's the invisible layer that gets cleaned.
Steps to clean your AI image with Calabi
- Upload your AI-generated image or video file to calabilabs.com.
- Calabi automatically strips all AI detection signals and injects a phone device profile.
- Review the forensic proof card showing the before/after metadata breakdown.
- Download your cleaned file and upload it to your platform of choice.
Frequently asked questions
Can Calabi remove visible watermarks like the Sora sparkle or a tool's corner logo? No — Calabi does not edit pixels or remove visible marks. If you need to remove a visible watermark, cropping works for corner logos. Calabi removes the invisible detection metadata that survives cropping, which is what actually gets you flagged after you've already cropped.
Will platforms still detect my image as AI-generated after cleaning? Calabi removes the metadata signals and encoder fingerprints that automated scanners check. Results vary by platform and by how the source model embedded invisible watermarks in the pixel data itself. Calabi addresses the metadata layer fully — that's the most common detection vector.
Does this work on video files too? Yes. Calabi handles video through the same three-stage pipeline — stripping C2PA and encoder signals from the bitstream, injecting phone capture metadata, and providing a forensic proof card. The Lavc and x264 SEI fingerprints in video files are specifically targeted.
Try Calabi free at calabilabs.com — 10 cleans, no card.