demorphus
Removing ordinary song tags does not establish that an AI watermark or audio fingerprint has been removed. Metadata, provenance credentials and audio classification are different layers. Suno’s own credentials checker looks for C2PA information; it explicitly is not a general-purpose AI music detector.
Three checks that answer different questions
| Layer | Question | What a negative result does not prove |
|---|---|---|
| Ordinary tags | What title, artist or encoder information is present? | That the audio was made by a human |
| Content Credentials | Does the file carry supported provenance information? | That no AI system was used |
| AI music classifier | How does this model classify this audio? | That every provenance mechanism is absent |
“Watermark remover” is used loosely in product searches. Before evaluating a claim, identify the signal being tested and the tool capable of testing it. The Suno watermarks guide explains the broader terminology.
What Suno officially documents
Suno describes Content Credentials as C2PA metadata attached to newly downloaded songs in its rollout, rather than a change to how a song sounds. Its checker detects those credentials and is explicitly separate from general AI detection. The announcement excludes songs downloaded before the rollout.
Therefore, a credentials result and an acoustic classification need not agree. Preserve the exact download and the report instead of treating one service as a substitute for the other. Suno’s page documents its credentials system; it does not establish the absence of every other possible identification method.
Why a Suno label from SH Labs is a different observation
A classifier identifying a likely generator is not, by itself, evidence that it decoded a deliberate watermark. To claim removal of a specific watermark, you would need to identify that mechanism and use an appropriate verification method. A before-and-after change in a classifier’s output supports the narrower statement that its prediction changed.
SH Labs describes its predictions as probabilistic. Keep the service, date and exact report with the file. Our AI music fingerprints guide explains why generator attribution and proof of provenance should stay distinct.
Check your export without guessing what changed
- Archive the original download and its creation details.
- Inspect the ordinary tags and note the tool used.
- If credentials are relevant, check them using a compatible verifier and retain the result.
- Run the audio checker separately when you need a classification report.
- For a processed export, repeat the same checks and compare sound independently.
A changed filename, file hash or visible tag list cannot establish which audio cues remain. A score also cannot certify compliance with a platform’s upload requirements.
Choose processing based on what you need to measure
Demorphus targets AI fingerprints and checks the processed file with an external detector. Compare Cleaner plans for processing, or use the free AI music checker for a baseline. For audible defects, begin with Suno artifact repair.
Check the source service’s terms before changing provenance information. Suno’s current terms restrict changes intended to conceal or misrepresent provenance. A detector result does not change the track’s creation history or the permissions attached to it.
Sources reviewed October 30, 2026. Repair workflows are suggested listening comparisons, not additional product benchmarks.
Prepared with AI assistance. Sources below.