Calabi Labs · Guide · 2026-06-14
Ai image editor
What "AI Image Editor" Actually Means in 2026
If you typed "ai image editor" into Google, you're probably looking for one of two things: a tool that uses AI to edit photos (think object removal, background replacement, retouching), or a way to handle AI-generated content so it doesn't get flagged, stripped, or shadowbanned on social platforms. Those are completely different problems — and most of what shows up in search results only solves the first one. Calabi solves the second one. It's not a photo editor in the traditional sense — it doesn't change how your image looks. It changes what your file reveals about itself at the metadata and encoding layer, which is exactly what platforms like Instagram, TikTok, YouTube, and Reddit are actually scanning.
What Actually Gets Your AI Image Flagged
When you upload an image, platforms don't read your mind — they read your file. Specifically, they scan for a set of invisible signals that have nothing to do with what's visible in the picture. Here's what's actually in a typical AI-generated export:
- C2PA / Content Credentials (JUMBF manifests): Cryptographic manifests embedded in the file that say, in structured data, "this was made by an AI model." Adobe, Microsoft, and Google are all signatories. If your image has a C2PA manifest, it lists the exact model, software version, and generation parameters. That's the "Made with AI" label you see on some images in Google Search and YouTube Shorts.
- XMP DigitalSourceType: trainedAlgorithmicMedia: An XMP metadata tag that explicitly flags an image as output from a trained AI model. This isn't subtle — it's a structured claim baked into the file by tools like Firefly, DALL-E, Midjourney, and Stable Diffusion exports.
- Encoder fingerprints: AI video exports especially carry recognizable encoder fingerprints in the bitstream. SEI (Supplemental Enhancement Information) NAL units from Lavc (FFmpeg's libavcodec) and x264 are dead giveaways. A file encoded with "Lavc" and then re-exported still carries that fingerprint in the stream headers.
- Missing capture signals: A real phone photo has GPS coordinates, a capture timestamp, a device Make/Model, and software version in the EXIF. AI exports typically have none of these, or they have generic placeholder values. Platforms treat missing capture metadata as a soft signal of non-camera origin.
A raw AI export from Midjourney or DALL-E can carry 144 metadata tags. Most of them mean nothing to a human eye — but automated content moderation systems read every single one.
Why Cropping, Screenshots, and Re-Uploading Don't Work
Here's what people try first when their AI image gets flagged:
- Cropping: Removes the visible frame and any visible watermark — this is the only real benefit. But all the metadata survives. C2PA manifests, XMP tags, and encoder fingerprints are embedded in the file's data structure, not in the pixel region you're cropping. A platform scanner reads the full file, not just the visible area.
- Screenshots: Re-captures the visible content but often preserves the metadata if the screenshot tool preserves EXIF — and critically, a screenshot from an AI image display interface still carries "screen capture" signals that platforms can use for detection.
- Re-exporting through a photo editor: Most desktop editors strip some metadata but don't know to target C2PA manifests or XMP AI flags specifically. A tool like Photoshop will strip camera serial numbers but won't touch the trainedAlgorithmicMedia tag.
- Re-uploading to "compress" the file: Platforms like Instagram and TikTok recompress on upload, which does strip some metadata — but C2PA manifests can survive recompression in some formats, and the platform's own scanner may have already read the metadata before processing.
None of these approaches target the specific signals platforms are actually scanning for. That's the gap Calabi fills.
How Calabi Actually Cleans an AI Image or Video
Calabi is a one-pass web tool. You upload a file, it runs automatically, and you download the cleaned result with a forensic proof card showing exactly what changed. Here's what the pipeline does:
- Strip detection signals: Calabi removes all C2PA / JUMBF manifests, zeroing out every Content Credentials atom. It strips XMP DigitalSourceType: trainedAlgorithmicMedia tags and any generator/tool metadata — the fields that explicitly say "AI-made." It targets Lavc and x264 SEI fingerprints in video bitstreams and removes encoder-specific fingerprints from image headers.
- Inject authentic phone-capture identity: Calabi writes real phone metadata into the file: a device profile (iPhone 15 Pro, Pixel 8 Pro, Galaxy S24 Ultra), a capture timestamp, GPS coordinates, and a known-phone encoder name. This replaces the "no capture metadata" signal that platforms flag as suspicious.
- Return a forensic proof card: Before you download, you see an ExifTool readout — the same forensic scan that newsrooms, platforms, and courts use — showing exactly which fields were stripped and which were injected. You can verify it yourself.
The result is a file that looks, to an automated scanner, like it came from a real phone — because the metadata now says it did.
FAQ
Does Calabi erase visible watermarks or logos?
No. Calabi doesn't change any visible pixels. If there's a visible watermark in the corner of your image, cropping removes it — and that's a legitimate approach for that specific problem. Calabi handles the invisible layer (metadata, encoding signals) that survives cropping and that platforms scan for automatically.
Will this guarantee my post won't get flagged?
No tool can guarantee that. Platform detection systems evolve constantly and vary by provider. Calabi removes the structured metadata and encoding signals that automated scanners specifically target — which is what causes preemptive removals and restrictions. A human-reported post or a system that relies on perceptual hashing (rather than metadata) is a different problem entirely.
What's the difference between Calabi and a photo editor like Photoshop or GIMP?
Photo editors change what you see — they let you paint over, clone, inpaint, and reconstruct pixels. Calabi changes what your file says — it works on the metadata, encoding fingerprints, and cryptographic manifests that sit underneath the image data. You need both for different problems. If your problem is a visible logo, use a photo editor. If your problem is an invisible "AI-made" flag in your file metadata, use Calabi.
Does Calabi work on video?
Yes. The same encoder fingerprint issue (Lavc, x264 SEI) applies to AI-generated video exports. Calabi's pipeline targets those signals in video bitstreams the same way it handles image metadata.
What's included in the forensic proof card?
It's a full ExifTool readout of the file before and after processing. You'll see JUMBF/C2PA atom counts reduced to 0, DigitalSourceType flags removed, metadata tag counts reduced from 144 to around 94 neutral structural tags, and new device profile fields injected. You can run the same scan yourself on any ExifTool-capable system.
Try Calabi free at calabilabs.com — 10 cleans, no card.