Local tool
PNG Info online
Drop in an image made with Automatic1111, Forge or ComfyUI to see its saved prompt, negative prompt, seed, steps, sampler and model: the same text the PNG Info tab shows, without opening the web UI. PNG, JPEG and WebP are read in your browser and not uploaded. Saved text is a claim, not proof of who made the picture.
Updated 25 October 2026 · Check AI Free
Automatic1111’s web UI documents a PNG Info feature that adds generation parameters to a PNG as a text chunk, so later software that can view PNG chunks can read them. When the web UI or Forge saves a JPEG or WebP instead, the same infotext goes into the EXIF UserComment field, and this reader looks there too. ComfyUI’s workflow-metadata docs say a PNG from a built-in save node can carry two JSON fields, prompt and workflow, written as PNG text. Either field can be missing. A file saved with metadata disabled, or rewritten by another app, will not yield the graph. This page reads those chunks. It does not launch ComfyUI, and it does not decide that the image is authentic.
Local PNG text reader
Read saved prompts and workflow fields
Choose an original PNG, JPEG or WebP. Prompt, negative prompt, seed, steps, sampler and model are shown from PNG text chunks or the EXIF UserComment that A1111 and Forge write to JPEG and WebP. The file stays in this browser and is not uploaded. It reads existing fields; it does not authenticate them.
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. This page only reads saved text. It does not run the image checker and does not upload the file.
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.
Local A1111 saved-parameters reader
Read prompts and generation settings
Paste saved A1111 parameters text. Positive and negative prompts, seed, steps, CFG, sampler and model fields are separated without running a generation. Unknown settings and the complete original text remain available.
Project-authored saved-parameters fixture. Prompts, model names, hashes and extension settings are fictional. PNG text limits: 32 fields, 256 KiB per field, 1 MiB decoded total. Pasted parameters: 1 MiB and 512 settings. Recognized settings require at least three key/value pairs with a generation-setting anchor; JSON-quoted strings and balanced parentheses/brackets/braces retain inner commas.
Parsing reference: official A1111 infotext utilities at 82a973c04367. This reader interprets supported saved syntax and preserves unknown values; it does not apply upstream defaults, styles, scripts or extensions. Files and text are not uploaded.
What the reader is doing
The same component runs on the Stable Diffusion page. Here you can inspect saved parameters and download the original text. The separate ComfyUI page puts its workflow summary first. It checks the PNG signature, chunk boundaries, CRC, and the text encodings used by tEXt, zTXt and iTXt. Latin-1 text, UTF-8 text and zlib-compressed text are in scope. Text stored after the image data is kept. Unknown chunks are skipped after their integrity check. For JPEG and WebP it reads the EXIF UserComment, decoding the ASCII and UNICODE (UTF-16) forms, and labels the field “parameters” only when it contains a Steps setting. Pixels are not decoded. The W3C PNG specification describes these text chunks; the reader follows that container, not a private Stable Diffusion format.
Limits, enforced before anything is rendered as HTML: 10 MiB per file, 4,096 chunks, 32 text fields, 256 KiB for a single field and 1 MiB of decoded text in total. JSON walks are capped at 10,000 values and a depth of 40, and a workflow nodes array at 1,000 entries. A browser without DecompressionStream gets an explicit message for compressed chunks. Text is shown as text. You can copy a field or download valid JSON unchanged. Review it before you share it. Prompts and workflows often contain private paths, names and unpublished wording.
Read a result without inventing the missing half
Step 1
Use the original file, not a screenshot
A screenshot, or an image re-saved by an editor or a chat app, usually loses the saved text. The reader will say when nothing is stored. If the picture passed through a messenger, read the metadata guide before you expect a workflow to have survived.
Step 2
Find the keyword, not just a long string
ComfyUI documents prompt and workflow. Automatic1111’s wiki describes generation parameters stored as PNG text. Of three public Automatic1111-related PNGs checked against Pillow for this project, two carried a tEXt keyword parameters. The reader prints the keyword it found. It does not rename fields to match a tutorial.
Step 3
Treat a parameters block as a claim
A checkpoint name, a seed and a prompt describe what the file says. They can be copied onto another image. They do not identify the person who clicked “generate,” and they do not survive every export. If the field is missing, it stays missing. One of the public reference PNGs used while checking this reader had a large compressed profile and no parameters field at all.
Step 4
Do not run the JSON because it parsed
A ComfyUI workflow from an image can reference nodes you do not have. Parsing it here only means the text fit the reader’s limits and, when it is JSON, that it was bounded well enough to flag as JSON. Open it in ComfyUI only if you trust the source and you expect to review every node.
Step 5
Ask a different tool for a different question
Want a general estimate of whether the pixels look generated? Use the image checker or the checker on the Stable Diffusion page. That path can send a smaller copy to a detection provider. This reader does not. Want the manifest, the claim generator and the recorded actions? Use the C2PA checker. It does not verify the signature. A workflow is not a credential. The node list for a ComfyUI graph is on the workflow page.
Why PNG Info shows nothing
An empty result almost always means the text was never saved or was removed later. The common causes: the image was generated with metadata saving turned off; it was downloaded from a site or app that strips metadata (most social networks and messengers do); it is a screenshot or was re-exported by an editor; or it came from a generator that does not write infotext at all, which includes most hosted services. None of these can be undone from the pixels. If the image came from Civitai or a similar gallery, try the original download rather than a preview or thumbnail.
What was checked, and what was not
Format tests cover truncated files, bad CRCs, bad UTF-8, unsupported compression, oversized text and literal markup that must stay text. Separately, three unchanged PNGs from Wikimedia Commons were read with Pillow 12.3.0 and with this reader. Field names, text and dimensions matched. Two are public-domain files associated with Automatic1111 in their Commons descriptions; the third is CC BY-SA 4.0 by Benlisquare and also described there as an Automatic1111-front-end image. Those are interoperability checks on historical public files. They are not a survey of current ComfyUI versions, and they are not a user study. The ControlNet reference was subsequently opened in Chromium: the displayed parameters could be copied and downloaded, with all 508 characters matching the stored field. The official ComfyUI SDXL example also displayed its two fields and downloaded API prompt JSON unchanged. Browser acceptance of the other two historical A1111-related references remains unclaimed. The external originals are not stored in this repository. JPEG and WebP reading is tested on files written with Pillow and piexif the same way the Automatic1111 web UI saves them, plus hand-built little-endian and ASCII variants. It has not yet been checked against a broad set of real downloads.
If the PNG info is empty, the picture can still be generated, photographic or both. Continue with how to tell if an image is AI-generated instead of treating silence as a result.
Sources
- Automatic1111 features wiki — Documents PNG Info as generation parameters stored in a PNG text chunk.
- ComfyUI workflow metadata — Field names, PNG text storage, how to disable metadata, and the warning that a workflow is not a signature.
- ComfyUI Save Image — The built-in node can embed workflow metadata when metadata saving is enabled.
- W3C PNG specification, text chunks — tEXt, zTXt and iTXt as defined for PNG.
- ControlNet Canny example on Wikimedia Commons — One of the public PNGs whose text matched Pillow during the reader check. See the file page for the license.
Related
- ComfyUI workflow from an image — The same reader, with the workflow summary first.
- EXIF viewer — Camera, GPS, XMP and IPTC in JPEG, PNG and WebP.
- C2PA checker — A manifest reader. Signatures are not verified.
- Stable Diffusion image check — The same reader, plus the general source-clue and model check.
- Check an image — For the pixel question this reader does not answer.