AI image detector by platform

Pixiv AI image check: the AI-generated setting and the original file

Pixiv asks creators to mark a work as “AI-generated” when posting and shows that label on the work, but the label is a self-declaration that lives on the page, so the file itself must be checked for any C2PA manifest, XMP statement or saved prompt.

Code-drawn illustration for Pixiv

Illustration · PNG file

  1. Signature + IHDR
  2. iTXt XML:com.adobe.xmprecord
  3. tEXt parameters
  4. tEXt prompt / workflow
  5. tEXt invokeai_metadata
  6. eXIf
  7. caBX C2PA manifest
  8. IDAT pixels
  9. IEND

AI source type declared

The file labels itself as AI-made.

IPTC DigitalSourceType such as trainedAlgorithmicMedia is read from XMP and shown with its exact value.

Check an image →
AI source type declared: where this record sits inside a PNG file and what the isGenAI report says about it. The picture is a code-drawn illustration.

What Pixiv strips and adds

Pixiv’s help center describes an “AI-generated” option at the bottom of the posting and editing pages for illustrations, manga, Ugoira and novels, for works where all or most of the production was done by AI. Marked works carry a label, appear in separate AI-generated rankings and can be hidden by viewers who choose to hide AI-generated work in search or across pixiv. Pixiv announced in 2022 that past works tagged #AI-generated would be marked automatically.

We found no pixiv documentation, as of October 2026, of what metadata it keeps or strips, or of C2PA support. As widely reported, the original-size image pixiv offers can still contain the PNG text chunks a generator wrote, which is why browser extensions that display prompts on pixiv only work when metadata survives. The smaller preview images are resized copies.

What isGenAI shows for a Pixiv image

The AI image detector reads the file you upload: a C2PA manifest if one survived, the IPTC DigitalSourceType in XMP, AI statements in metadata, saved prompts in the AUTOMATIC1111, ComfyUI and InvokeAI formats, and matching copies by hash and perceptual fingerprint. A platform copy with none of these returns “no evidence found”, and the report says plainly that absence proves nothing about the author.

Download the original-size image rather than saving the preview: if the creator uploaded an unedited PNG, it is the copy most likely to still hold a saved prompt.

What to do instead

Note whether the work carries pixiv’s AI-generated label and keep the work URL. Check the original-size file, not a screenshot or a resized repost. When the file returns no evidence, ask the creator for their generator’s original export.

A work without the label is not verified as hand-made: the setting depends on the creator’s answer.

How to check an image from this platform

  1. Note the platform’s own label or credential icon on the post before you download anything; it does not travel with the file.
  2. Ask the poster for the original file, or trace the first upload. A platform download is a re-encoded copy on most services.
  3. Upload what you have to the AI image detector and read the report. A stripped copy returns “no evidence found” honestly.
  4. Use the matching-copy check: a hash or perceptual fingerprint match can tie the stripped copy to an original that still carried records.

Check the downloaded image

Frequently asked questions

How does pixiv label AI-generated works?

Creators set the “AI-generated” option when posting or editing. Marked works show a label, join AI-generated rankings and can be hidden by viewers, per pixiv’s help center.

Can I hide AI-generated works on pixiv?

Yes. Pixiv lets you hide AI-generated works in search results or choose “Hide” for AI-generated work in user settings to remove them across pixiv.

Does pixiv keep the prompt in the image?

Pixiv does not document it. As widely reported, an original-size PNG can keep the generator’s text chunks; upload it to isGenAI to see what survived.

Does pixiv support Content Credentials?

We found no pixiv documentation of C2PA Content Credentials as of October 2026.

Sources

Where a platform documents its behaviour, the vendor page is linked. Where it does not, the page says the behaviour is reported, not documented.