FAQ · AI content & watermarks
Sometimes — a re-encode disrupts many, but robust watermarks (like SynthID-style signals) can persist; the reliable wins are the metadata and encoder signals.
Sometimes, but not always — this is the honest hard case. Some AI tools embed invisible pixel-domain patterns that survive JPEG compression, cropping and screenshots. A controlled re-encode with light noise disrupts many of them, and aggressive modes add perspective, scale and crop shifts that defeat perceptual-hash matching, but robust watermarks (such as SynthID-style signals) can persist. That is why the reliable wins are the metadata and encoder signals you can fully remove — C2PA, generator tags, DigitalSourceType flags and encoder fingerprints. Strip those completely, inject a real phone identity, and you’ve closed the signals detection actually relies on most.
Related on Calabi: Does Instagram detect AI images? · How do I remove C2PA metadata? · Does cropping remove a watermark? · How to post AI content on Instagram without getting shadow-banned · Sora — or just open the tool.