Reply to thread

Message: [QUOTE="Bombastus, post: 92218, member: 2178"] ElevenLabs lets a Professional Voice Clone owner share a voice only with named workspace members instead of exposing it publicly. Permission can therefore be enforced at the account layer before anyone reaches the generation controls. Consent permits a voice to exist, but distribution is a separate decision. A system built around [B][URL='https://goldmidi.com/community/threads/personalized-vocals-do-not-mean-artist-voice-cloning.77087/']personalized vocals without unrestricted artist cloning[/URL][/B] can limit who gets the voice, which accounts can operate it, and which requests meet platform rules. Those are product controls, not merely clauses sitting in a contract. [HEADING=2]Consent does not have to mean open access[/HEADING] Professional Voice Clones can stay inside a workspace, be shared with selected users or groups, go out through an external sharing link, or appear in the public Voice Library. Each route creates a different audience before the voice generates anything new. The external link is deliberately broad. Anyone with the link can add an eligible Professional Voice Clone to an ElevenLabs account, while a workspace owner can instead restrict the same resource to named users or groups. An artist can therefore authorize a model without making the model publicly discoverable or generally available to unrelated users. Workspace permissions go further than a simple yes-or-no share. ElevenLabs supports admin, editor, commenter, and viewer roles for shared workspace resources, including voices. A team can give different people different levels of control without duplicating the voice or handing every collaborator the same authority. Service accounts can be fenced in as well. ElevenLabs documents them as workspace principals that start with no access to workspace resources, then receive access only when an administrator shares a resource with them. A production API process does not have to inherit everything a human user can see. This matters for licensed voices because a contract often names a project, team, client, or approved operator. Software cannot interpret every clause automatically, but access rules can still make some violations harder by ensuring the voice never appears in accounts that were not supposed to use it. [HEADING=2]Moderation can stop some uses at request time[/HEADING] Distribution controls decide who reaches the voice. Live Moderation deals with a different problem by checking requested text against prohibited content categories when an eligible Voice Library voice has the feature enabled. Access and permitted use are separate. Permission to reach a voice does not automatically mean permission to make it say anything, and a performer may accept one commercial use while rejecting another. Platform screening can block certain request categories before audio is returned, rather than relying entirely on somebody noticing misuse afterward. ElevenLabs also warns that Live Moderation can add latency. A safeguard applied during generation therefore has an operational cost, however small it may be in a particular deployment, and products using a moderated voice need to allow for it instead of treating moderation as invisible. Current controls still stop short of turning a private licensing agreement into executable policy. ElevenLabs describes Live Moderation around its prohibited categories, not as a custom rule engine where an artist can enter every approved brand, territory, script, release date, and campaign exception. A 2025 [B][URL='https://patents.google.com/patent/US20250029614A1/en']synthetic-speech authorization watermark patent[/URL][/B] describes a more granular approach in which machine-readable authorization data can specify who may use audio, how it may be used, and whether particular purposes are allowed. The patent does not describe ElevenLabs’ implementation, but it shows how usage permission can become a technical object rather than a note somebody has to remember. [HEADING=2]API access can be narrower than team access[/HEADING] Workspace controls matter even more once generation moves from a browser into an application. ElevenLabs can share a voice resource with a user, group, or service-account API key, so an automated process can receive access to one resource without being given blanket workspace visibility. A resource can currently be shared with up to 100 principals. The ceiling is mundane, but it forces a real design choice for larger operations between individual assignments, group-based access, and dedicated service accounts instead of treating one permanent credential as the universal solution. Public sharing behaves differently. Professional Voice Clones can be shared externally with anyone holding the link, while Instant Voice Clones and Voice Design voices are limited to workspace sharing. Choosing the wrong sharing route can therefore widen the audience far beyond the people named in a licensing workflow. Withdrawal adds one final boundary. Voice Library owners can use a notice period before previously saved access ends, while a voice without one can disappear immediately when sharing stops. New users cannot keep adding a removed listing during an active notice period. For API users, an active notice period can expose a disable_at_unix value, and ElevenLabs offers voice-removal webhooks for larger deployments. A product can use those signals to revoke routes, swap an approved replacement, or stop new jobs before the voice reaches its disable date. [/QUOTE]

Name