AudioSeal and ElevenLabs on two different senses of open
Both entries concern audio and both describe a way to detect a mark. One is open to anybody willing to build it; the other is open to anybody with an account, and only for files that vendor made. From 2 source pages, read 2026-09-22; no maker pays to be listed.
| Point | Meta AudioSeal | ElevenLabs |
|---|---|---|
| What is available | A published scheme with a detector described in the paper | A working detector inside the product |
| Open to whom | Anybody willing to implement it | Anybody with access, for this vendor's marks |
| What it will not do | Tell you which products use the scheme | Find another platform's watermark |
Inclusion rule. Two entries whose published answers to one column pull in different directions. Cells hold each vendor's own wording from the page its entry names; pairs that agree on the column are not filed at all. Order. Fixed order, and the two columns are in the order the heading gives them.
1Open to implement and open to use are not interchangeable
A platform with engineers can turn a paper into a service. A journalist holding a clip cannot, and will not, and needs something that already exists behind a web form.
Both entries are described as open and they serve those two readers in opposite proportions. The record keeps the reading route as a row rather than a verdict for exactly this reason.
2Audio schemes are not interchangeable, so neither are detectors
A detector has to know what it is looking for. Nothing obliges competitors to publish their schemes, so cross-platform checking needs an agreement rather than a better detector.
That is the gap the codes-of-practice paragraph anticipates for this whole subject. Until something fills it, a per-vendor route is the only route that actually runs, and a published method is the only thing anybody else can build on.
3Neither claim reaches a delivered mix
One measures a quality trade-off on the material it was tested against. The other describes a tool for finished files from one product. A generated line mixed under music and compressed for a platform is outside both.
That case is where nearly all published audio actually lives. The pair marks the boundary rather than guessing past it, because guessing past it would mean inventing a measurement.
- Meta AudioSealA perceptual loss inspired by auditory masking, presented as achieving better imperceptibilitythe stated quality trade-off
- ElevenLabsThe detector can only find watermarks embedded by this vendor and will not find those of other platformsthe reach of the checking route
4A provider's own account
Against the duty in “AudioSeal and ElevenLabs”, SceneMixer's published pages describe what its pipeline produces:
Dubbing replaces the audio of a finished clip; the picture was generated for a different line, so mouth shapes, pauses and shot length all belong to the original language. Native-language generation gives the video model the line in the target language before it renders the shot, so the performance, the pacing and the cut points are built for that language.
SceneMixer, Native-Language Dialogue vs. Dubbing in AI Video (2026): Why the Cast Should Speak the Line (read 2026-10-13)
SceneMixer writes dialogue into the script analysis, assigns each line to a shot with a speaker and a short delivery note, and the video model performs it — picture and audio generated together, so the mouth forms that sentence in the language you chose.
SceneMixer, AI Dialogue Generator (read 2026-10-13)
5Sources
Read from the research publication at ai.meta.com and the documentation at elevenlabs.io, consulted 2026-09-22. Full entries: Meta AudioSeal and ElevenLabs. The column itself is on Reading route. Other pairs on file: Sora and Nova Reel, SynthID and Synthesia.