BrowserTools
Home / AI / AI Image Metadata Viewer

AI Image Metadata Viewer

See the prompt and settings saved inside an AI-generated image, check whether it carries an AI label, and remove that data if you want.

Maintained by BrowserTools · Last updated 2026-09-29

Loading AI Image Metadata Viewer… If nothing happens, please enable JavaScript.

Most AI image generators quietly write their recipe into the file they save. AUTOMATIC1111, Forge and SD.Next put the full prompt, negative prompt, sampler, steps, CFG scale, seed, size and model name into a PNG text chunk called parameters. ComfyUI goes further and embeds the entire node graph twice, once as the executable prompt and once as the editable workflow. NovelAI stores its settings as JSON in a Comment chunk, InvokeAI uses invokeai_metadata, and Midjourney leaves the prompt with its --ar and --v flags in the image description. Drop an image here and this tool finds that data, recognises which generator wrote it, and lays it out as readable fields with a Copy button next to the prompt and the negative prompt, much like the PNG Info tab in the Stable Diffusion web UI.

Frequently asked questions

Is my image uploaded to read its metadata?
No. The file is read as raw bytes inside your browser tab and every chunk and segment is parsed by JavaScript on the page. Nothing is sent to a server, which matters when the prompt, the workflow or the file name is something you would rather keep to yourself.
Which AI generators does it recognise?
AUTOMATIC1111, Forge and SD.Next (the parameters text), ComfyUI (the prompt and workflow graphs), NovelAI (the Comment JSON plus Software and Source), InvokeAI (invokeai_metadata and the older sd-metadata and Dream keys), Fooocus and Draw Things style JSON, and Midjourney descriptions with their flags and Job ID. Anything else that stores text in the file still shows up in the raw metadata viewer, so you can read it even when the format has no dedicated parser.
How do I get the ComfyUI workflow out of an image?
Drop the PNG here and, if it was saved by ComfyUI with metadata enabled, a Download workflow.json button appears under the parsed prompt. You can drag that file onto the ComfyUI canvas to rebuild the graph. Dragging the original PNG onto ComfyUI works too; the JSON file is handy when you want to version the workflow, share it without the image, or inspect it in an editor. When only the API-format prompt is embedded, the tool offers workflow-api.json instead.
Why does my image show no metadata when I know it was AI generated?
Because the data rarely survives sharing. Discord, X, Instagram, Facebook, WhatsApp and most image hosts re-encode uploads and drop text chunks, EXIF and XMP. Screenshots, copying from the browser and many editors do the same. Some web UIs also have a setting to save images without parameters. An empty result therefore tells you nothing about how the image was made.
Can this tool tell whether any image is AI generated?
It can only report what the file says about itself. Generation parameters, an IPTC digital source type of trainedAlgorithmicMedia or a C2PA manifest from an AI service are strong signals, but they can be removed in seconds and could in principle be added to a real photo. It does not analyse pixels and cannot see invisible watermarks such as Google SynthID, which need the vendor's own detector.
What is C2PA and what does manifest present mean?
C2PA is the standard behind Content Credentials. A signed manifest describing who created or edited the file is stored inside it, in a PNG caBX chunk, in JPEG APP11 segments or in a WebP C2PA chunk. This tool detects that manifest and reads the claim generator, any digital source type and the certificate organisation. It does not check the signature or the hash of the pixels, so for a verified answer upload the file to the official Content Credentials verify site.
What does IPTC DigitalSourceType trainedAlgorithmicMedia mean?
It is a value from the IPTC NewsCodes vocabulary that says the image was created by a model trained on sampled content, which is to say generative AI. compositeWithTrainedAlgorithmicMedia means a real image was edited with generative AI, as with Photoshop Generative Fill. Other values such as digitalCapture declare an ordinary camera photo. Firefly, Google, Meta and several stock libraries write this field so that platforms can show an AI label.
How does Remove metadata work, and does it reduce quality?
For PNG, JPEG and WebP the default mode is lossless: the tool copies the image data untouched and leaves out text chunks, EXIF, XMP, comments and C2PA blocks, while keeping the colour profile. The pixels are identical to the original. A second option redraws the image through a canvas, which removes absolutely everything but applies a new compression pass. If a JPEG relied on an EXIF rotation tag, use the re-encode option so the orientation is baked in.
Why is the prompt in my JPEG shown correctly here but garbled elsewhere?
AUTOMATIC1111 saves JPEG and WebP parameters in the EXIF UserComment field with a UNICODE prefix followed by UTF-16 text. Many EXIF viewers either ignore that tag or decode it with the wrong byte order, which produces a string of odd characters. This tool detects the byte order from the data itself before decoding.

About AI Image Metadata Viewer

For ComfyUI images the tool follows the links in the graph from each sampler back to the text encoders, so you see the positive and negative prompt even when the text comes through a primitive or a combine node. It lists the checkpoint, diffusion model, VAE, text encoders and every LoRA with its strength, shows the node count, and offers the workflow as a workflow.json download you can drag straight back onto the ComfyUI canvas. JPEG and WebP files are covered as well: the EXIF UserComment and ImageDescription fields where web UIs store parameters are decoded, including the UTF-16 variant that trips up most EXIF readers, and so are the prompt and workflow tags that ComfyUI writes into animated WebP files.

The second question people ask about an image is whether it was made by AI at all. Adobe Firefly, Google, Meta, OpenAI and a growing list of cameras and editors now declare the origin of a picture in two standard places: the IPTC DigitalSourceType field in XMP, where the value trainedAlgorithmicMedia means generative AI, and a C2PA Content Credentials manifest stored in a JUMBF box inside the file. This tool reports both, reads the claim generator and the certificate organisation out of the manifest, and says plainly what that means: a manifest is a signed claim this page does not cryptographically verify, and missing metadata proves nothing because social networks, messaging apps and screenshots strip it. Everything runs locally with no upload, and a Remove metadata button strips the text chunks, EXIF, XMP and C2PA blocks losslessly when you want to share an image without its prompt.

PNG Info was a happy accident

The PNG format has allowed free text inside images since its first specification in 1996. The tEXt chunk was meant for modest things like a title, an author or a copyright line, and for twenty-five years almost nobody looked at it. That changed in 2022, when the AUTOMATIC1111 web UI started saving the whole generation recipe into a chunk named parameters, and its PNG Info tab turned every shared image into a reproducible experiment.

ComfyUI took the idea to its logical end by embedding the complete node graph, so a single PNG can carry dozens of kilobytes of JSON describing every loader, sampler and upscaler used to make it. Communities formed around swapping workflows as images, and sites such as Civitai built galleries where the prompt is read straight out of the file.

The same years pushed the industry toward the opposite kind of metadata: labels meant to disclose AI rather than share it. The C2PA coalition, founded in 2021 by Adobe, Microsoft, Intel, the BBC and others, defined a signed manifest that travels inside the image, and the IPTC extended its digital source type vocabulary with terms such as trainedAlgorithmicMedia so that software could say, in a standard way, that a picture came from a generative model. Both depend on the file keeping its metadata, which is exactly what most social networks still throw away.

Related tools

EXIF Viewer & Remover

See the hidden data in a photo, like the camera and GPS location, and save a copy without it.

Image to Text (OCR)

Copy the text out of screenshots, scans and photos, right in your browser.

Image Compressor

Make images smaller in file size and compare the result with the original.

Audio to Text

Turn a recording or video into text and subtitles, right in your browser, without uploading it anywhere.
Support