Calabi Labs · Guide · 2026-06-17
How to remove the midjourney watermark
You can crop out Midjourney's visible watermark, and most people stop there. What they miss is the invisible metadata layer — C2PA Content Credentials, XMP AI flags, and encoder fingerprints — that platforms scan for automatically, often within seconds of upload. Here's what actually happens, why cropping only solves half the problem, and how to clean the other half.
What actually gets your Midjourney image flagged
Midjourney embeds two separate detection systems into every image it exports. The first is visible: a small "MAGICAL" text marker and a job ID string, typically positioned near the lower edge of the image. The second is invisible, and this is what trips up creators on Instagram, TikTok, Reddit, and YouTube.
That invisible layer includes JUMBF / C2PA Content Credentials — cryptographic manifests stored directly in the image file that declare the image was AI-generated. Midjourney was one of the first generators to adopt this standard. It also embeds an XMP DigitalSourceType tag set to trainedAlgorithmicMedia, an IPTC-approved field that flags AI-trained content. A raw Midjourney export carries roughly 144 metadata tags. Even after you crop out the visible mark, those invisible tags remain intact inside the file.
Platforms in 2026 scan for C2PA atoms, XMP AI flags, and encoder fingerprints like Lavc (FFmpeg's encoder) or x264 SEI in video exports. Missing GPS, capture timestamp, and device identity also register as synthetic signals. You can't see these with your eyes, but the platform's automated systems read every single one.
Why cropping and screenshots don't solve the metadata problem
Cropping works for the visible watermark — you cut it out and it's gone from the image. But the metadata is embedded in the file structure itself, not in any visible region. When you crop a Midjourney image in Photoshop or Preview, you're changing the pixel dimensions while leaving the underlying EXIF, XMP, and JUMBF data blocks untouched. The platform still sees a file full of AI-generation signals, just with fewer pixels.
Screenshotting is even less effective. Taking a screen recording of a Midjourney image preserves the original file's metadata and adds a new capture layer on top. The original AI-generation signals are still in the file you uploaded — now buried under a screen capture wrapper.
Re-exporting through a tool like Photoshop or FFmpeg can strip some metadata, but it rarely touches C2PA JUMBF atoms specifically, and it typically doesn't address the encoder fingerprint that identifies the generation pipeline. You might go from 144 tags to 80 and still have the critical AI-signature tags intact.
How to fully clean a Midjourney image for upload
Calabi runs a single automatic pipeline that strips the invisible detection layer and injects authentic phone-capture identity. Here's what the process looks like:
- Upload your Midjourney image. Drag and drop the file — no settings to configure, no tool to select.
- The strip stage runs automatically. Calabi removes all JUMBF / C2PA Content Credentials atoms, zeroes out the DigitalSourceType: trainedAlgorithmicMedia XMP flag, strips generator/tool tags, and clears encoder fingerprints like Lavc and x264 SEI headers.
- The inject stage runs automatically. Calabi writes a fresh device identity — Make, Model, Software version, GPS coordinates, and capture timestamp — from a real phone profile (iPhone 15 Pro, Pixel 8 Pro, Galaxy S24 Ultra, and others available).
- Review the forensic proof card. Before downloading, you see an ExifTool readout — the same forensic scan platforms use — showing exactly what was stripped (all AI-generation signals) and what was injected (phone-capture identity). This is what "clean" looks like under the hood.
- Download the cleaned file. The output is a standard JPEG or PNG with neutral structural metadata — no AI flags, no generator fingerprints, and a phone-capture identity profile.
Frequently asked questions
Will cropping alone pass platform detection?
For the visible watermark, yes — cropping removes the pixel region where "MAGICAL" or the job ID text appears. For the invisible metadata layer, no. Platforms flag files based on embedded metadata, not visual content, so the watermark removal and the metadata cleaning are separate problems requiring separate solutions.
Can I just use EXIF metadata strippers instead?
Generic EXIF strippers remove some tags but don't reliably target C2PA JUMBF atoms or XMP AI flags like DigitalSourceType. They also don't inject a replacement device identity, which means your file goes from "flagged as AI" to "no camera identity at all" — which is itself a synthetic signal platforms can detect.
What about Midjourney's paid plans — do they remove watermarks?
Midjourney's subscription tiers affect image quality, resolution, and remix options. Neither the standard nor the Pro plan removes the visible watermark from exported images, and none of them strip the C2PA or XMP metadata. The metadata problem is a platform-level issue, not a Midjourney account-tier issue.
Try Calabi free at calabilabs.com — 10 cleans, no card.