Image Metadata, EXIF & AI Tag Viewer

Drop any image to instantly see its hidden data — camera and EXIF info, GPS location, and whether it was made by AI. For AI images, it reveals the embedded prompt and generation settings (Stable Diffusion, ComfyUI, NovelAI and more). Everything runs in your browser; your image is never uploaded.

Detect AI-generated images & read their prompt

Most AI image tools quietly store how a picture was made inside the file. Stable Diffusion (AUTOMATIC1111), ComfyUI, NovelAI and InvokeAI write the full prompt, negative prompt, model, seed, sampler, steps and CFG scale into PNG text chunks. Images from DALL·E, Adobe Firefly and others often carry C2PA "content credentials". This tool reads all of that locally and shows you the prompt and parameters — handy for learning how an image was made, or just checking whether something is AI-generated.

What this tool shows

• AI generation — detects the generator and shows the embedded prompt, model, seed, sampler, steps and CFG when present.
• File details — name, size, type and last-modified date.
• Image details — pixel dimensions, aspect ratio and megapixels.
• EXIF — camera make and model, lens, ISO, aperture, shutter speed, focal length and date taken.
• GPS — latitude and longitude, with a link to view the spot on a map.

Privacy first

Everything is read in your browser — the file is never uploaded. If you find GPS, a prompt, or other data you'd rather not share, you can strip it by re-saving the image: run it through our image compressor or convert it (for example PNG to JPG), which produces a clean copy without the original metadata.

What EXIF actually is

EXIF is a block of structured data written into the image file next to the pixels — not something derived from them. In a JPG it sits in an APP1 segment near the start of the file: a small TIFF-style directory of numbered tags, each holding one value such as camera make, exposure time, orientation or a GPS coordinate. The decoder that draws the picture on screen skips over it entirely.

That separation explains the two behaviours people find surprising. Copying, renaming, emailing or backing up a file carries the EXIF along untouched, because the bytes travel as one unit and nobody has to understand them. But anything that decodes the image to pixels and encodes it again — resizing, a filter, a screenshot, most "save as" operations — is writing a brand new file, and unless the tool deliberately copies the old block across, the metadata is simply not there any more.

Which formats carry which metadata

• JPG — the richest. Camera make and model, lens, ISO, aperture, shutter, focal length, timestamp, GPS, and often an XMP block with editing history on top.
• HEIC — the iPhone default. Same EXIF tags as JPG, plus Apple-specific fields for things like depth data and Live Photos.
• PNG — no standard camera EXIF. What PNG has is text chunks (tEXt, iTXt, zTXt), free-form key/value pairs, and that is exactly where AI generators put their prompts. Some tools do jam EXIF into a PNG anyway, with mixed results depending on the reader.
• WEBP and AVIF — can hold EXIF and XMP in optional chunks, but whether they do depends entirely on what produced the file. Web-optimised copies usually don't.
• TIFF — EXIF is built on the TIFF directory structure, so support here is native.
• GIF — nothing useful beyond a comment extension.
• Screenshots — dimensions and maybe a software tag. There was no camera, so there is nothing to record.

What each generator writes

AUTOMATIC1111 puts everything in one tEXt chunk keyed parameters: the prompt on the first line, Negative prompt: on the second, then a comma-separated tail with Steps, Sampler, CFG scale, Seed, Size, Model hash and Model. It is plain text and reads like a settings dump, because that is what it is.

ComfyUI takes a different approach and writes two chunks, prompt and workflow, both JSON, holding the entire node graph — every node, its class type, its inputs, and how they are wired together. Far longer and much less pleasant to read, but it is a complete recipe: drop that PNG back into ComfyUI and the graph reassembles itself. NovelAI uses Software, Comment and Source keys; InvokeAI writes invokeai_metadata, or sd-metadata and Dream on older builds.

C2PA content credentials are a different animal — a cryptographically signed manifest embedded by Adobe apps, OpenAI's image tools and a few camera makers, describing what created the file and what was done to it. Its presence does not by itself mean an image is AI-generated; it means something in the chain chose to sign its work.

None of this is proof. Metadata can be stripped in seconds and it can be forged. But when it is present it usually tells you far more than staring at the picture does — a prompt, a seed and a model hash is a specific, checkable claim about where a file came from, and hunting for six-fingered hands is not.

Checking before you share

The images worth checking are the ordinary ones: photos taken at home or at work, pictures attached to a marketplace listing, anything going to someone you don't know, anything about to be posted in public. GPS is the obvious risk — a phone photo taken in your living room pins your address to roughly a metre — but the timestamp and camera serial number matter too, since they quietly link images that otherwise have nothing to do with each other.

Big platforms re-encode uploads and strip most metadata in the process, which is real but should not be relied on. Direct file transfers keep everything: email attachments, files sent as documents rather than photos in chat apps, cloud drive links, most forum and marketplace uploads. And a site removing EXIF from the public copy says nothing about what it kept on its own servers.

To strip metadata yourself, re-encode the file. Running it through the image compressor at high quality, or converting it with PNG to JPG, rebuilds the image from pixels alone, so nothing from the original block survives into the copy.

Frequently asked questions

How do I see an image's metadata?
Drop the file into the viewer above. It reads the file details and any EXIF present — camera model, settings, date taken, GPS if it is there — and shows them straight away. Nothing is uploaded.

Can I read the prompt from an AI image?
Frequently, yes. Stable Diffusion via AUTOMATIC1111, ComfyUI, NovelAI and InvokeAI all store the prompt and generation settings inside the PNG. This tool pulls those chunks out and shows the prompt, model, seed, sampler and step count when they are present.

Can a photo give away where I live?
It can. Phone cameras embed GPS coordinates by default in many setups, accurate enough to identify a specific building. Checking here before you send a file is the point of the tool.

Does my image get uploaded?
No. Parsing happens in JavaScript on your own machine and the file never leaves it, which is why inspecting private or sensitive photos here is safe.