Private AI Image Detector That Runs Locally

Most AI image detectors are cloud services: you upload the image, a model on their servers scores it, and you trust them with the file. For private photos, client work or sensitive material, local checks are the safer first step.

What can run locally

Metadata checks don't need a server. Many AI tools leave traces in the file, such as generator names, Stable Diffusion prompts and settings, IPTC tags marking AI content, and Content Credentials.

• AI Image Detector: AI Image Detector (made by Patrick Bushe, who runs this site) reads these in your browser when you right-click an image in Chrome. Its standard check doesn't send the image to Bushe.co or any detection service. It stores only your settings and a scan count.
• ExifTool: a free, open-source desktop program that reads every metadata field in a file, fully offline.

The trade-off

Local metadata checks only work when metadata is present. Social media images, screenshots and re-saved files often have none, and then a local check can only say "No Metadata Found."

Visual detection, which analyzes the pixels, needs large trained models. Commercial ones such as Hive and SightEngine run on their servers. AI Image Detector's optional Deep Scan uses your own key for one of those services, so it does upload the image; keep it off if privacy comes first.

Running a model yourself

Researchers have published open-source AI image detection models you can run on your own computer with some technical setup. They're often trained on older generators and may not recognize images from newer ones, so test them on examples first.

Other private checks

• Visual inspection: zoom in on hands, text and reflections.
• Context: who sent it, and can they show the original?
• Cropping before uploading: if you must use a cloud service, crop out faces or details you don't want to share.

See checking images without uploading.