AI image detector by platform
X AI image detector: what a saved post image can prove
X publishes no documentation of what it strips, and files saved from X posts are widely reported to arrive without EXIF, XMP or C2PA records, so an image from X is checked by tracing the original upload and using isGenAI’s matching-copy check rather than by reading the download.

Illustration · PNG file
- Signature + IHDR
- iTXt XML:com.adobe.xmprecord
- tEXt parameters
- tEXt prompt / workflow
- tEXt invokeai_metadata
- eXIf
- caBX C2PA manifest
- IDAT pixels
- 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 →What X strips and adds
We found no X help page describing metadata handling or an AI label policy as of October 2026. Independent reports consistently describe EXIF being removed from uploaded images, and the images X serves are re-encoded at several sizes. Grok images generated inside X circulate through the same pipeline, so a Grok image saved from a post is normally stripped as well.
isGenAI measured it on 19 October 2026: we uploaded five test files through the X API and downloaded the served copies. X removed the C2PA manifest, the IPTC AI source type, the EXIF camera and GPS fields, the AUTOMATIC1111 saved prompt and the ComfyUI workflow from every file, while keeping the format, dimensions and decoded pixels. The files, hashes and method are in our platform metadata survival study under /research.
Because nothing is documented, treat X as a platform that preserves nothing until a specific file proves otherwise.
What isGenAI shows for an X download
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.
Save the highest-resolution variant X offers; it is still a re-encode, but it gives the perceptual fingerprint the best chance of matching a known copy.
What to do instead
Ask the poster for the original file, or find the first publication elsewhere. Reverse-tracing a viral image usually leads to a source that still carried a credential or a saved prompt. The Grok generator page covers images made inside X.
Keep the post URL with your check so a reviewer can retrace it.
How to check an image from this platform
- Note the platform’s own label or credential icon on the post before you download anything; it does not travel with the file.
- Ask the poster for the original file, or trace the first upload. A platform download is a re-encoded copy on most services.
- Upload what you have to the AI image detector and read the report. A stripped copy returns “no evidence found” honestly.
- Use the matching-copy check: a hash or perceptual fingerprint match can tie the stripped copy to an original that still carried records.
Frequently asked questions
Does X remove metadata from images?
Yes, in our 19 October 2026 test X removed C2PA, IPTC/XMP, EXIF/GPS, saved prompts and ComfyUI workflows from every test file. X publishes no policy. Assume a stripped copy.
Does X label AI-generated images?
We found no documented X policy on AI labels as of October 2026. Do not rely on the presence or absence of a label.
Can isGenAI detect AI in an image saved from X?
Only from surviving records or a matching copy in its records. Most X downloads return no evidence found; trace the original.
How do I check a Grok image seen on X?
Ask for the export from the Grok interface and check that; the X copy is normally stripped. The Grok page explains what to expect.
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.