Leonardo AI check: reading the descriptive fields
Leonardo output is worth reading the way you would read any platform that leaves descriptive metadata rather than a signature: check the XMP and EXIF fields for a tool or model name, and for video output walk the container as well. AICheck365 does that locally in your browser and reports each hit with the field it came from, so you can judge how much weight it deserves rather than taking a verdict on faith.
Short answer: expect unsigned tool fields, corroborate with provenance
Read the original export for tool strings in XMP and EXIF — Software, CreatorTool, Credit and any model identifier — and, for video, the MP4/MOV container fields. These are readable and often specific, but unsigned, which means they can be edited, copied onto another file, or lost the first time the file is re-encoded. Treat a hit as a supporting clue checked against where the file actually came from.
Sample status and evidence profile
| Bundled Leonardo sample | Not bundled yet. This page does not claim a reproduced Leonardo test until a licensed original can be published alongside the other samples. |
|---|---|
| What the parser matches | The tool-pattern matcher recognises Leonardo strings in metadata field values and in file names, and reports which field produced the match. |
| Evidence level | Unsigned descriptive metadata: weaker than a verifiable C2PA manifest, and reported as a lower-confidence supporting signal. |
Signals to Check
| XMP / EXIF fields | CreatorTool, Software, Credit and model identifiers can name the generating stack when the original export survives intact. |
|---|---|
| Model strings | Named models sometimes appear in the descriptive fields, which helps distinguish pipelines within the same platform. |
| Video container metadata | For video output, ftyp, the box tree and udta/meta/ilst entries carry encoder and tool fields the same way as any other MP4. |
| File name | Download names can carry a platform pattern or job identifier. Context only — renaming removes it. |
Unsigned does not mean useless
It is easy to move from "this metadata can be faked" to "this metadata is worthless", and that leap is wrong. Forging a plausible tool field requires someone to have deliberately done it; the ordinary case is that the field is either present and honest, or absent because a re-encode stripped it. What unsigned really means is that the field cannot stand alone against an adversary — so it is read alongside where the file came from, who supplied it, and whether the other fields agree with it.
When to escalate to a signed check
If the question is consequential — a dispute, a publication decision, anything with a cost attached — descriptive fields are a starting point rather than an answer. The escalation is to look for a C2PA manifest, either on this file or on a copy obtained closer to the source, because that is the only layer whose integrity can be verified rather than trusted. When neither is available, the honest report is that the file no longer supports a firm conclusion.
Limits
- Leonardo signals are unsigned metadata: they can be edited, removed or copied onto an unrelated file.
- Social platforms, messaging apps, screenshots and re-exports routinely strip XMP and EXIF fields entirely.
- No bundled Leonardo sample is published yet, so this page describes field coverage rather than a reproduced test result.
- No signals found means this copy carries no readable evidence — it does not indicate the file is not from Leonardo.
Use the EXIF/XMP reader for the descriptive fields; run the full detector when you also want C2PA and PNG parameters checked.
FAQ
How do I check whether an image came from Leonardo?
Read the original export's XMP and EXIF fields for a tool or model name. The check is free, runs in your browser without uploading the file, and reports each hit with the field it came from.
Does Leonardo add C2PA Content Credentials?
Do not assume so. The dependable signals here are descriptive metadata fields. If a signed manifest appears, it was most likely added by a later editing step rather than at generation.
Can it be identified after posting to social media?
Often not from the file alone. Platforms re-encode uploads and drop XMP and EXIF data, leaving nothing readable to report. The original export is far more informative.
Are files uploaded for checking?
No. Parsing runs locally in your browser, so the file never leaves your device.
Sources and verification records
- AICheck365 public sample records — Which samples are bundled, their provenance, test dates and known limits.
- IPTC Digital Source Type vocabulary — Definitions of controlled values such as trainedAlgorithmicMedia.
- AI image metadata guide — How the descriptive and signed layers are weighted against each other.
- AICheck365 methodology — How unsigned metadata is reported and rated.