View the prompt and metadata of an AI image

How do I see the prompt of an AI image?

Drop the file above, click to choose it, or paste it. A short verdict comes first, then “What the image contained”, which opens with the prompt and generation settings. Each text is shown under the name of its key, with who wrote it when the key is known: parameters for AUTOMATIC1111, Forge or Fooocus, prompt and workflow for ComfyUI, invokeai_metadata for InvokeAI.

Long texts get their own box with a Copy button, which copies the full text even when the page shows only the first 20,000 characters. JSON, such as a ComfyUI workflow, and texts over 2,000 characters start folded: click to open them. PNG, JPEG and WebP files up to 50 MB are read; AVIF, HEIC and GIF are not supported yet, and the tool tells you so.

What does each part of the report mean?

• Prompt and generation settings: the text blocks of a PNG, such as the A1111 infotext (prompt, negative prompt, seed, sampler, model and LoRA hashes) or a ComfyUI workflow. When A1111 saves a JPEG or WebP, the same text sits in EXIF instead, under “User comment”.
• EXIF: camera maker and model, software, dates, description, and the GPS position as latitude and longitude when there is one. A maker note or an embedded thumbnail is listed with its size only.
• XMP: the tool that created the file and its digital source type. The IPTC vocabulary uses trainedAlgorithmicMedia for “created using generative AI” and compositeWithTrainedAlgorithmicMedia for an image edited with it, such as an inpainted photo. A description, a credit or a link to Content Credentials can appear here too.
• Content Credentials (C2PA): how many manifests the file holds and their size, the tool it says it was made with, and the signer as declared. PromptWipe does not check the signature; the Content Authenticity Initiative’s Verify site does.
• Other data: an HDR gain map, secondary images such as a depth map, the short video of a Motion Photo, or data hidden after the end of the image.

Below come “What was removed” and “What was kept”, then a folded list of every block of the file in order, with its position and size, for the curious.

Can I read a ComfyUI workflow without opening ComfyUI?

Yes. ComfyUI stores two JSON texts in its PNG files: prompt, the graph as it ran, and workflow, the graph as you see it on the canvas. The report shows both, folded, with the number of nodes in the workflow. Copy them into a text editor to read the node settings, the model and LoRA file names, and every prompt in the graph. The page on removing the ComfyUI workflow explains what it reveals.

Does viewing the metadata change my file?

No. Your browser reads a copy of the file in memory; the file on your disk is never touched. PromptWipe also prepares a cleaned copy and offers it as a separate download, named after your file with -clean added. If you only wanted to read, ignore it. Nothing is stored: close the tab and the image is gone from the page.

Why does my image show no metadata?

Usually because it was removed before the image reached you. Common reasons:

• Social networks and messaging apps often re-encode uploaded images and drop their metadata on the way. A file saved from such a site has often lost its prompt already.
• A pasted image often loses its metadata on the way to the clipboard. Drop or choose the original file instead.
• A screenshot carries none of the original file’s metadata.
• The tool did not write any. A1111 has a setting to turn its infotext off, ComfyUI has a server option that stops it saving the workflow, and Fooocus saves metadata only when you turn it on.

An empty report does not mean the image was not made with AI. Invisible watermarks such as SynthID are part of the pixels, not of the metadata, and PromptWipe neither detects nor removes them.

Is it the same as the PNG Info tab?

Same idea, different place. AUTOMATIC1111’s PNG Info tab reads the infotext inside the web UI and can send the settings back to txt2img. PromptWipe needs nothing installed and also names ComfyUI, InvokeAI and Fooocus keys, lists EXIF fields, XMP and Content Credentials, and shows every block of the file. When you decide to share the image, the same page removes all of it: see remove metadata from AI images and the page on Stable Diffusion metadata.

Is my image uploaded to read it?

No. The reading happens in the page itself. The site’s security policy forbids the page from sending data anywhere, so the tool keeps working once the page is loaded, even offline. You can check it in your browser’s developer tools: the network tab stays empty while the image is read. How it works and the privacy page say more.

To see which tools write which fields, read the guide to AI image metadata. For phone photos, see photo metadata and GPS; for provenance data, C2PA Content Credentials.

Sources: IPTC Digital Source Type vocabulary; C2PA 2.2 specification; AUTOMATIC1111 modules/images.py; ComfyUI nodes.py and comfy/cli_args.py.