Calabi Labs · Guide · 2026-06-19
Clothes wrinkle remover
If you generated a clothing product shot with AI and it keeps getting rejected or labeled as AI-generated on Instagram, TikTok, or the platform you're posting to, the problem isn't the wrinkles in the fabric. It's the invisible metadata layer buried in your file that platforms scan for before your post even goes live.
What actually gets your clothing photo flagged
When you export an AI-generated image — whether from Midjourney, Firefly, Stable Diffusion, or any other tool — the file carries an invisible forensic trail that platform scanners detect in seconds. This has nothing to do with how the clothes look. It's structured data embedded in the file itself.
The primary signal is the C2PA / Content Credentials manifest, stored as JUMBF metadata atoms. This is a cryptographic manifest that explicitly states the image was generated by an AI model — not captured by a camera. It travels with the file even after you crop it or convert it to a different format. Beyond that, XMP metadata tags like DigitalSourceType: trainedAlgorithmicMedia flag the file as AI-derived at the metadata level. Video exports add another layer: encoder fingerprints like Lavc (FFmpeg's libavcodec) and x264 SEI messages in the bitstream are dead giveaways of synthetic generation. Platforms also scan for the absence of expected phone-capture signals — no GPS coordinates, no capture timestamp in EXIF, no real device maker/model. A raw AI export reads as structurally anomalous compared to a real phone photo.
In short: the wrinkle-free clothing looks fine. The file's metadata signature says "AI" before a human ever sees it.
Why cropping, screenshotting, and re-exporting don't work
Screenshotting a clothing photo and re-exporting it strips some visible quality but leaves the C2PA and XMP metadata intact in many cases — especially if you're screenshotting from an AI tool's preview that already burned a visible watermark into the pixels. Cropping removes the visible watermark and improves composition, but the cryptographic manifest and XMP AI flags survive the crop and get re-encoded into the new file. Re-exporting from Photoshop or Preview does rename some fields, but it doesn't strip the C2PA manifest or the trainedAlgorithmicMedia flag — those require deliberate forensic-level removal. What you're left with is a visually cleaner image carrying the same AI provenance invisible to you but readable by every platform scanner.
How to actually clean an AI-generated clothing photo
Calabi is a one-pass web tool that strips the detection signals and injects authentic phone-capture identity so your file reads as a normal phone photo at the forensic level. Here's how it works:
- Upload your AI-generated clothing image. Drop the file — PNG, JPG, WebP, or video. No manual settings.
- Automatic strip and inject runs. Calabi removes every C2PA / Content Credentials JUMBF manifest, every XMP DigitalSourceType flag, every encoder fingerprint (Lavc, x264 SEI), and the full AI metadata layer — 144 metadata tags in a raw AI export get reduced to roughly 94 neutral structural tags. Then it injects a real device profile: iPhone 15 Pro, Pixel 8 Pro, or Galaxy S24 Ultra, with authentic Make, Model, Software, GPS coordinates, and capture timestamp.
- Review the forensic proof card. Before download, you see the exact same ExifTool scan that platform scanners use — what was stripped, what was injected. You verify it worked, not guess.
- Download the cleaned file. Post it normally. The file-level identity now matches a real phone capture.
FAQ
Does Calabi remove visible wrinkles or logos from clothing photos?
No. Calabi does not edit pixels, use inpainting, or remove visible elements from images. If your AI clothing photo has a visible watermark or logo in the corner, crop it out before uploading — Calabi handles the invisible metadata and detection signals that survive cropping.
Can I use Calabi for clothing product videos as well as still images?
Yes. Calabi processes both images and video files. The pipeline strips C2PA manifests and encoder fingerprints from video bitstreams the same way it handles still image metadata.
Will this guarantee my clothing post won't get flagged?
No tool can guarantee a platform won't flag any upload — platform detection systems evolve and vary. Calabi removes the metadata, XMP, and encoder signals that automated scanners check. Results vary by platform and by the source model used to generate your image.
Try Calabi free at calabilabs.com — 10 cleans, no card.