Skip to content
checkaifree
AI image detector

Stable Diffusion image checker

Check if an image is AI-generated, and read PNG text when the file still has it. Automatic1111 parameters and ComfyUI workflows are clues about what someone saved, not a signature.

Need saved settings? Read PNG text locally ↓
Checking scope
Files
JPEG, PNG, WebP · up to 10 MB
Model check
General AI-image estimate, when available
On this page
PNG parameters and ComfyUI text
Signatures
Not verified by this site
Specimen plateAwaiting image

Drop an image here

choose a file

Metadata demos (controlled fields):
JPEG · PNG · WebPUp to 10 MBNo sign-up

Local PNG text reader

Read saved prompts and workflow fields

Choose your original PNG to inspect tEXt, iTXt and zTXt fields, including text stored after image data. This reader stays in your browser and does not upload the file. It reads existing fields; it does not authenticate them or run a workflow.

No file chosen, or drop one here

Synthetic sample: project-authored stored text, not a real generator output. Up to 10 MiB; 32 text fields; 256 KiB per field and 1 MiB total decoded text. Workflow JSON is bounded. Image pixels are not decoded by this reader. Other detector inputs on this page use the separate processing described above.

ComfyUI/A1111 fields may be absent after export, screenshotting or editing. Text can include private prompts or paths: review it before sharing. Downloaded JSON is the stored field, not a verified or safe-to-execute workflow.

How the file travels

Local SD files are metadata-rich until they aren’t

Stable Diffusion is a local habit more than a single model. SD 1.5, SDXL, SD3 and a pile of fine-tunes share the same export customs. Automatic1111 writes a parameters text chunk: prompt, negative prompt, sampler, seed, size, model hash. ComfyUI can write prompt and workflow JSON. Forge and other forks follow one of those patterns or none.

The chunk sits in the PNG. Save as JPEG, screenshot the UI, or upload to a site that recompresses, and the text is gone. Turning metadata off in the UI means it was never there. The reader on this page walks PNG text chunks in the browser, including compressed text, and does not upload the file for that read. It cannot rebuild a workflow that is not in the file, and it does not run the graph.

There is no default C2PA manifest and no SynthID on a typical local install. A user can add either with a separate tool. A checkpoint name inside a workflow is a claim the workflow makes. Our general classifier does not identify the checkpoint, the LoRA, or the seed.

The drawings are original illustrations of file paths. They are not outputs from Stable Diffusion.

Diagram of a Stable Diffusion PNG with parameters versus a JPEG export without themSD PNGJPEG exportText removed
The parameters live in the PNG. The JPEG of the same render is a different, poorer file.
Stable Diffusion PNG layers: parameters, workflow JSON, pixelsA1111 parameters chunkComfyUI prompt and workflowNo default C2PA or SynthID
A1111 and ComfyUI write different text. Both are editable. Neither is C2PA.
What the Stable Diffusion page reads locally versus the general modelPNG readerModel checkPrompt textWorkflow JSONNot a checkpoint IDNot a signature
The PNG reader stays in the browser. The optional model check is a separate, general score.
Metadata and credentials

Metadata and Content Credentials

The useful credential here is plain PNG text, which anyone can edit. A parameters block that names a model hash is still just text. Compare it with the creator’s files if you know them. Do not treat it as a camera, and do not treat a missing block as a camera either.

Hosted SD services sometimes strip those chunks on purpose and sometimes add their own watermark. Read what this file contains. The first-64KB source sniff used by the general check is not the PNG reader: the reader is the separate control on this page, and it is the one that sees text after the pixel data.

This site can notice source text in the first part of a file. It does not verify C2PA signatures, asset bindings or certificate trust, and it has no SynthID decoder.

Official check

There is no Stable Diffusion verifier to send the file to

A typical local install writes no C2PA manifest and no SynthID watermark. The practical “official” record is the text the UI saved: Automatic1111’s parameters chunk, or ComfyUI’s prompt and workflow JSON. That text is the creator’s note. It can be edited, deleted, or never written. The reader on this page shows the chunks that are still in the PNG. It stays in the browser, and it does not run the graph.

A checkpoint name inside that text is a claim, not a signature. Hosted Stable Diffusion sites sometimes strip the chunks and sometimes add their own watermark. Read this file. Do not import a Flux or OpenAI credential story onto it.

Look twice, then don’t conclude

Visual tells, treated cautiously

There is no Stable Diffusion look that survives every checkpoint and LoRA. Pretending there is will mislabel photographs and other generators.

Checkpoint-dependent faces

One SDXL fine-tune is glossy. Another is grimy. The tell is the checkpoint the PNG names, if it names one, not a universal texture.

Sampler debris

Old SD 1.5 images showed repeating textures and fused objects. Current fine-tunes were trained to hide that. Its absence means nothing.

UI screenshots

Sometimes you were sent a screenshot of the Automatic1111 window, parameters and all, as pixels. That is visible context. It is not embedded metadata, and the reader will not OCR it.

How this check handles the file

Source clues are checked in your browser first. When needed, file-header data and a smaller image copy may be sent to our backend and detection provider. Model copies have embedded metadata removed. Image URLs are retrieved by our server; see the privacy policy for processing and provider retention. Privacy details

Sources for the generator notes

Frequently Asked Questions

Questions about Stable Diffusion.

The answers below are about this generator’s files, not a promise to name the model.

How do I read Stable Diffusion PNG info?

Use the PNG text reader on this page with the original PNG. It shows parameters, prompt and workflow text when those chunks exist. It does not recover them from a JPEG or a screenshot, and it does not upload the file to do the read.

Can the text prove who made the image?

No. PNG text is editable. It can support a creator’s account of their settings. It is not a signature and not an identity check.

Will the detector name the checkpoint?

No. The optional score is a general AI-image estimate. Checkpoint, LoRA and seed come from the text, if the text is there, and even then only as a claim.

Is the check free?

No account is required. The local PNG read does not use the model quota. Model checks still depend on availability.

Explore Other Generators

Also inspect other generative models

All checking guides
Image review scenarios

Reviewing profile photos or art entries?

Review LinkedIn candidate photos or competition artwork with context, and read our method and limitations.

Method and limits