AI Image Metadata Cleaner

Generators write their fingerprint into the file: Stable Diffusion WebUI and ComfyUI store the full prompt and model name in PNG text chunks, NovelAI and InvokeAI tag themselves by name, and Firefly, DALL·E, and Google's models attach C2PA Content Credentials. The tool reads those blocks and matches them against known signatures.

No. Metadata is trivially removed, and screenshots, re-saves, and most social platforms strip it automatically. A clean result means there is no signature left in the file — not that the image is real. Invisible pixel watermarks like Google's SynthID cannot be read without the vendor's detector.

Not at all. The file is rebuilt by copying the original image data byte for byte and dropping only the metadata blocks — there is no re-encoding, so the pixels are bit-identical to the original.

PNG text chunks (tEXt, zTXt, iTXt), EXIF, XMP, IPTC, timestamps, and C2PA Content Credentials. Colour-critical data — ICC profiles, the Adobe colour-transform marker, palettes, and animation frames — is deliberately kept so the image still displays correctly.

No — and that's the point. Other footprint removers upload your file to a server and promise to delete it later, usually within some number of minutes. This tool parses and rebuilds the file inside your browser, so there is no upload, no queue, no retention window, and no file-size cap. Your prompt never reaches anyone.

Yes. Drop as many PNG, JPG, and WebP files as you like — each one is scanned separately, and you can download them individually or all at once as a ZIP.