Calabi Labs · Guide · 2026-06-14
Ai image combiner
Here is the full page, written as clean HTML, targeting the "ai image combiner" query with honest framing and Calabi's specific capabilities.
```html
What "AI Image Combiner" Actually Means — and What Happens to Your File
An AI image combiner is a tool that takes two or more images and uses an AI model to merge, blend, or composite them into a single output — automatically matching lighting, edges, perspective, and style. Tools like PhotoGrid, Visualgpt, ImageCombinerAI, and HeadshotMaster do this in one click. The problem isn't the result looking fake to a human eye — the problem is that the output file arrives filled with invisible machine-readable signals that platforms scan and flag automatically.
When you download a merged AI image and try to post it on Instagram, TikTok, Reddit, or YouTube, it may get flagged, shadowbanned, or suppressed — not because the image looks wrong, but because the file's metadata layer screams "AI-generated." Calabi fixes that at the file level in one pass.
What Actually Gets Your File Flagged
Platforms don't flag images by looking at pixels — they scan the invisible metadata layer underneath. When an AI image combiner processes your photos, it embeds a specific chain of signals that automated systems are tuned to detect:
- C2PA / Content Credentials (JUMBF manifests): The AI combiner embeds a cryptographically signed manifest that states, in structured data, that the content was machine-generated. This is the same system Adobe, Microsoft, and the C2PA coalition have deployed across millions of AI export files. Platforms read this manifest directly. A typical AI-combined export carries 12–18 JUMBF atoms stating the file's AI origin.
- XMP DigitalSourceType: trainedAlgorithmicMedia: This specific XMP tag is the clearest signal an image came from a trained AI model. It's not a guess — it's a structured metadata field inserted by the generator. Removing it is one of the first things Calabi strips.
- Generator and tool metadata tags: The combiner software writes its name, version, and model details into the file's EXIF, IPTC, or XMP blocks. Fields like Software, CreatorTool, and model-specific tags from Stable Diffusion, DALL-E, Midjourney, or the combiner's own pipeline get embedded alongside your image data.
- Encoder fingerprints — Lavc, x264 SEI: Many AI image combiners export video or animated content through FFmpeg's Lavc encoder, which stamps the stream with Lavc parameters and x264 SEI (Supplemental Enhancement Information) messages. These are consistent across AI-generated content and form a recognizable fingerprint.
- Missing capture signals: A normal photo from an iPhone carries Make: Apple, Model: iPhone 15 Pro, GPS coordinates, and a capture timestamp in the EXIF. AI-generated or AI-combined files typically lack all of these. The absence itself is a signal platforms flag.
Why the Obvious Fixes Fail
If you've tried to post an AI-combined image and gotten flagged, shadowbanned, or had reach throttled, you may have tried one of these workarounds:
- Screenshotting the image: This removes the EXIF metadata layer, but platforms can still detect the AI generation through perceptual hashing — they compare your screenshot against known AI image fingerprints regardless of metadata. And screenshots introduce compression artifacts that themselves look suspicious to some classifiers.
- Cropping the edges: Removes the visible border but does nothing to the invisible metadata. The C2PA manifest, XMP tags, and encoder fingerprints survive cropping intact — they're embedded in the file structure, not the pixel boundaries.
- Re-exporting through another tool: Saving from Preview, Photoshop, or Canva strips some metadata, but C2PA manifests are designed to persist through re-encoding. Unless you specifically target the JUMBF atoms and XMP AI tags for removal, they travel with the file.
- Compressing or resizing: JPEG recompression disrupts some metadata but not the structured C2PA layer, and heavy recompression visibly degrades image quality — which itself can trigger quality-based suppression on platforms.
These methods treat the symptom (some metadata exists) without addressing the actual detection layer — the structured AI-signaling fields and encoder fingerprints that automated scanners read.
How to Actually Clean an AI Image Combiner Output
Calabi is a one-pass web tool that works on the file itself — stripping every detection signal and injecting authentic phone-capture identity before you download. Here's exactly what happens when you upload an AI-combined image:
- Upload your file. Drop the AI-combined image directly into Calabi at /app. No account needed to start.
- Automatic strip — the pipeline removes: All JUMBF / C2PA atoms (reducing C2PA references from double-digits to zero), the DigitalSourceType: trainedAlgorithmicMedia XMP flag, generator and tool metadata tags, Lavc and x264 SEI encoder fingerprints, and any other AI-signaling fields — leaving only neutral structural metadata.
- Automatic inject — authentic phone identity: Calabi writes real phone metadata into the file: a device profile (iPhone 15 Pro, Pixel 8 Pro, or Galaxy S24 Ultra), GPS coordinates, capture timestamp, and a genuine phone encoder name. The file now reads, byte-for-byte, as a normal phone recording.
- Forensic proof card: Before you download, Calabi shows you the ExifTool readout — the same forensic scan platforms use. It shows exactly what was stripped and what was injected. You see the clean result before you commit to it.
- Download the cleaned file. The file you save carries the authentic phone metadata profile and zero AI detection signals.
Before and After: What the Metadata Looks Like
| Field | AI-combined file (before Calabi) | After Calabi clean |
|---|---|---|
| C2PA / JUMBF atoms | 18 atoms present | 0 — fully stripped |
| DigitalSourceType XMP | trainedAlgorithmicMedia | Removed |
| Total metadata tags | ~144 AI-specific tags | ~94 neutral structural tags |
| Make / Model | Missing or AI-tool name | iPhone 15 Pro (or Pixel 8 Pro / Galaxy S24 Ultra) |
| GPS / Timestamp | Absent | Realistic coordinates + capture time |
| Encoder | Lavc / x264 SEI fingerprint | Authentic phone encoder |
FAQ
Does Calabi change how my combined image looks?
No. Calabi works entirely on the file's invisible metadata and encoding layer. The pixels — the actual merged image — stay exactly as you created it. If your AI combiner produced a smooth blend, that blend remains unchanged.
I used a visible watermark remover on my AI-combined image too. Does Calabi still help?
Yes — but for different reasons. Removing a visible logo or watermark with cropping or an inpainting tool doesn't touch the metadata layer. Platforms still detect the AI generation signals underneath, even after the visible mark is gone. Calabi removes the invisible detection layer that survives any visual editing.
Can I trust that platforms won't still detect my image?
No tool can guarantee a platform never flags you — detection methods evolve and some are proprietary. What Calabi does is remove the specific structured signals (C2PA, DigitalSourceType, encoder fingerprints) that automated scanners are explicitly tuned to find. Results vary by platform and source model, but removing these signals eliminates the most consistent and well-documented detection vector.
Try Calabi free at calabilabs.com — 10 cleans, no card.
```