AI image detector by generator
Stable Diffusion image detector: the prompt saved in the file
Stable Diffusion itself writes no provenance; the interface around it does, and isGenAI exposes the saved prompt, negative prompt, seed, steps and model name that AUTOMATIC1111, ComfyUI and InvokeAI store in the original PNG.
parameters-tEXt.pngPNG file · controlled example
- Signature + IHDR
- iTXt XML:com.adobe.xmp
- tEXt parametersrecord
- tEXt prompt / workflow
- tEXt invokeai_metadata
- eXIf
- caBX C2PA manifest
- IDAT pixels
- IEND
Saved prompt found
The generator wrote its settings into the file.
Prompt, negative prompt, seed, sampler and model are read word for word, with the field they came from.
Open the prompt extractor →Where the evidence comes from
Stable Diffusion is an open-weight model family, so what ends up in a file depends on the application that ran it. AUTOMATIC1111 adds the generation parameters to the PNG as a text chunk named parameters. ComfyUI saves its prompt graph and workflow in PNG text chunks. InvokeAI stores an invokeai_metadata record. None of these are signed; all of them are editable text.
Hosted services built on Stable Diffusion behave differently and often strip everything. As of October 2026 we found no vendor documentation of Content Credentials on Stability AI’s hosted outputs, so treat each service as unknown until you inspect a real file.
What isGenAI shows
The checker reads the parameters chunk, the ComfyUI prompt graph and the InvokeAI record and reports a saved generation prompt with the model name and settings when exactly one consistent record exists. If two records disagree or a multi-stage graph has several samplers, it reports a saved workflow instead of guessing which prompt produced the pixels.
JPEG and WebP exports from AUTOMATIC1111 keep the same text in EXIF UserComment, and isGenAI reads that too. The image prompt extractor shows the full saved prompt, and the ComfyUI workflow viewer displays the node inventory.
What a repost destroys
Any re-encode removes PNG text chunks: screenshots, format conversion, social uploads and most messaging apps. A Stable Diffusion image found on a feed therefore usually carries no prompt. The original PNG from the output folder is the file to check.
Because the record is plain text, it proves what a tool wrote, not who pressed the button. isGenAI labels it a declaration of moderate strength and keeps it separate from signed records.
How to check an image from this generator
- Get the original download, not a screenshot or a copy saved from a chat or social feed.
- Upload it to the free AI image detector. Files up to 20 MiB in PNG, JPEG, WebP, AVIF or TIFF are read on the server and are private by default.
- Open “How we checked” and read each finding with its source field: signed record, declared source type, saved prompt, AI statement or matching copy.
- Treat a clean result as “no evidence found”, never as proof that a person made the picture.
Frequently asked questions
Does Stable Diffusion add metadata to images?
The model does not; the interface does. AUTOMATIC1111 writes a parameters text chunk, ComfyUI writes prompt and workflow chunks, and InvokeAI writes invokeai_metadata. Hosted services vary and often strip metadata.
Can isGenAI show the Stable Diffusion prompt?
Yes, when the original PNG (or an A1111 JPEG/WebP with EXIF UserComment) still carries the record. The report shows the prompt, negative prompt, seed, steps, sampler and model name exactly as saved.
Is a Stable Diffusion image detectable after a screenshot?
Not from metadata. Screenshots and social uploads drop the text chunks, and isGenAI reports no evidence found rather than inferring from pixels.
Does a saved prompt prove the image is AI-generated?
It proves that a generation record was written into the file. The text is editable, so isGenAI reports it as a declaration of moderate strength, not as a signed proof.
Sources
Vendor behaviour changes. These pages describe what the vendor documents and what isGenAI reads from a file; they do not test every release.