AI image detector by generator

NovelAI metadata reader: prompt, seed and settings from the PNG

An original NovelAI PNG stores its generation record in two text chunks: Software set to “NovelAI” and a JSON Comment holding the prompt, the undesired content (uc), seed, steps, sampler, scale and size. isGenAI’s image prompt extractor reads that record directly from the file, free and without an account.

Controlled example PNG file that carries this record: Saved prompt found

parameters-tEXt.pngPNG file · controlled example

  1. Signature + IHDR
  2. iTXt XML:com.adobe.xmp
  3. tEXt parametersrecord
  4. tEXt prompt / workflow
  5. tEXt invokeai_metadata
  6. eXIf
  7. caBX C2PA manifest
  8. IDAT pixels
  9. 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 →
Saved prompt found: where this record sits inside a PNG file and what the isGenAI report says about it. The picture is a code-drawn controlled example that carries this record; download it and check it yourself.

What NovelAI writes into a PNG

NovelAI downloads carry plain PNG text chunks: Title, Description (the prompt), Software (“NovelAI”), Source (the model, for example NovelAI Diffusion V4.5) and Comment. The Comment chunk is JSON with the generation settings: prompt, uc (the negative prompt), steps, scale, seed, sampler, noise_schedule, width and height. V4 and V4.5 images also store the prompt as v4_prompt.caption.base_caption, with optional per-character captions.

NovelAI also hides a copy of the settings in the alpha channel in the “stealth pnginfo” format, described in NovelAI’s own metadata repository. That copy can survive tools that strip text chunks, but flattening to RGB or any lossy re-encode destroys it.

How to read NovelAI metadata with isGenAI

Open the image prompt extractor and drop in the original PNG you downloaded from NovelAI. isGenAI checks that Software says NovelAI, parses the Comment JSON and shows the prompt, negative prompt, seed, steps, sampler, scheduler, scale and size, each copied straight from the file. Nothing is guessed: if a value is missing it is left out.

When the V4 caption and the older prompt field disagree, or the image uses several character captions, there is no single saved prompt to report, so isGenAI says so instead of picking one. For a quick look at every raw chunk, including Title, Source and Generation time, use the EXIF viewer.

What isGenAI does not read yet

isGenAI does not decode the alpha-channel stealth pnginfo copy. If a NovelAI image has lost its text chunks but kept transparency, NovelAI’s own open tools are the right reader for it.

A screenshot, a JPEG conversion or an image re-saved by Discord, X or Instagram removes the Comment chunk, and flattening removes the stealth copy. For those copies isGenAI still checks Content Credentials, the IPTC source type, AI statements in metadata and matching copies, and reports “no evidence found” when nothing is left.

How to check an image from this generator

  1. Get the original download, not a screenshot or a copy saved from a chat or social feed.
  2. 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.
  3. Open “How we checked” and read each finding with its source field: signed record, declared source type, saved prompt, AI statement or matching copy.
  4. Treat a clean result as “no evidence found”, never as proof that a person made the picture.

View the raw fields in the EXIF viewer

Frequently asked questions

How do I see the prompt of a NovelAI image?

Upload the original NovelAI PNG to isGenAI’s free image prompt extractor. It reads the JSON Comment chunk and shows the prompt, negative prompt (uc), seed, steps, sampler, scale and size.

Where does NovelAI store its metadata?

In PNG text chunks (Description, Software, Source and a JSON Comment with the settings) and in a stealth pnginfo copy hidden in the alpha channel, which isGenAI does not decode.

Can isGenAI read NovelAI V4 and V4.5 images?

Yes, when the prompt is a single base caption. Images with several character captions have more than one prompt, so isGenAI reports that no single saved prompt can be recovered.

Why is no NovelAI prompt found?

The file is probably a screenshot, a JPEG or a copy re-saved by a chat or social app, which drops the Comment chunk. Use the original PNG download.

Sources

Vendor behaviour changes. These pages describe what the vendor documents and what isGenAI reads from a file; they do not test every release.