About PassReal

PassReal is an AI image humanizer. It takes an image made by a generator and makes the file behave like camera output.

What it does

Every image goes through four stages:

• Metadata stripping. PNG text chunks, XMP blocks and C2PA manifests are removed. On a Gemini export, 174 provenance markers go down to 0.
• Sensor simulation. Grain and radial lateral chromatic aberration, as a real lens and sensor produce.
• Frequency smoothing. A low-pass filter in the frequency domain softens generation residue.
• Camera encoding. A JPEG with camera EXIF and the quantization table of the declared body.

Accepted formats: PNG, JPEG and WebP, up to 20 MB and 8192 px on the longest side.

What it does not do

Naturalization changes what a file declares and what its pixel statistics look like. It does not change whether the subject of an image reads as machine-made to a person looking at it. Pixel watermarks such as SynthID are a separate layer, and PassReal does not claim to remove them.

How quality is measured

Fidelity is measured per image against your source, as SSIM and PSNR. Measured SSIM stays above 0.85. No detector pass rate is published: rates depend on the detector version, the generator and the subject, and any single figure would be out of date as soon as a detector is retrained. If a public detector flags an image you paid for, the money-back guarantee applies.

Your images

Results expire after 7 days and are then deleted. Your images are never used to train AI models. The details are in the Privacy Policy.

Contact

One address, read by a human: hello@passreal.com. See also the contact page and pricing.