Metadata · heuristics

AI image check

Inspect image files for metadata signals that indicate AI generation — EXIF software tags, creation tool fields, and format heuristics. Runs entirely in your browser.

Drop an image to analyze

Click or drag · one file, up to 12 MB

PNG · JPG · WebP · GIF · more

How AI Image Detection Works

This tool inspects image metadata — information embedded in the file itself — for patterns associated with AI image generation. It does not use neural network classifiers. Instead it looks for explicit signals in the file structure.

What signals are checked?

  • EXIF Software tag — AI generators like Midjourney, Stable Diffusion, DALL·E, and Firefly often write their name into the Software field.
  • PNG text chunks — Stable Diffusion embeds the full prompt, model checkpoint, sampler, and seed as PNG tEXt chunks.
  • XMP metadata — Adobe Firefly and some tools use XMP fields (GeneratedBy, CreatorTool) to declare AI involvement.
  • IPTC source fields — Some platforms embed AI attribution in IPTC Creator or Source fields.
  • Aspect ratio heuristics — Common AI output dimensions (512×512, 1024×1024, 768×512, etc.) score as a weak positive signal.
  • Format patterns — Certain combinations of bit depth, color profile, and absence of camera EXIF (Aperture, ISO, Shutter) suggest non-photographic origin.

Limitations

Metadata-based detection is easily defeated by stripping EXIF data before sharing. Many AI images have no identifying metadata at all, especially when downloaded from social media (which strips EXIF). This tool cannot detect AI content from pixel patterns alone — that requires a trained classifier model which this browser tool does not include.

Frequently Asked Questions

No. This tool checks for metadata signals only. A "likely AI" result means the file contains software tags or patterns consistent with known AI generators. A "clean" result means no such signals were found — not that the image is definitely real. Metadata can be stripped or forged.

It can detect them when the metadata has been preserved. Midjourney images downloaded directly from Discord retain EXIF Software metadata. Stable Diffusion images generated with Automatic1111 or ComfyUI embed detailed PNG text chunks. DALL·E images from the OpenAI interface often have no identifying metadata.

No. The entire analysis runs in your browser using the File API and binary parsing. The image bytes never leave your device. This makes it safe to check confidential or sensitive images.

JPEG (reads EXIF and IPTC), PNG (reads tEXt and iTXt chunks with XMP), WebP, GIF, and BMP. HEIC/AVIF metadata extraction depends on the browser's built-in decoder. RAW formats are not supported.