The new Fish Audio license demonstrates why a downloaded speech model should not be labeled simply “free for commercial use.” The threshold, hosted-service exception and consent rules all matter.
Separate small-business commercial use from hosting
The agreement dated 13 October 2026 offers limited commercial use below its stated annual gross-revenue threshold, but excludes Hosted Services without a separate written license. A studio generating its own narration and delivering finished audio is not necessarily treated the same as a self-serve public voice-cloning API. Read the contract definitions, including affiliates, outsourced customer deployments, model distribution and transition requirements after revenue grows.
Model names and dates are not interchangeable
The repository documentation distinguishes current community-license materials from model weights still carrying a separate research license. Do not retroactively apply the newest agreement to every historical checkpoint or state that every model is commercially cleared. Keep exact download checksum, model identifier, versioned license, code commit and redistribution record. If a chosen checkpoint has research-only terms, do not deploy it in a commercial product merely because newer documentation describes another model.
Voice consent and misuse restrictions
The license restricts generating recognizable voices without authorization or another lawful basis and prohibits deceptive impersonation or attempts to defeat provenance measures. Consent must be specific to voice samples, languages and use. For an internal branded narrator, prefer the voice of a voluntarily enrolled employee or a properly contracted performer. Even where model output ownership is assigned to the user, underlying recordings, script copyright and publicity rights are not erased.
Attribution, revenue and derivative distribution
The agreement includes attribution and notices for redistributing model materials and derivative works. A software vendor embedding the runtime has obligations different from a video producer distributing only a finished MP3. Organizations close to the threshold must design a compliance review rather than assuming continued eligibility. Legal counsel should interpret borderline cases; this is an editorial explanation, not a grant of license.
Decision framework
For private evaluation, compare hardware support, model fit and whether research use is permitted. For paid narration, validate the actual licensed checkpoint and output workflow. For SaaS voice generation, seek an explicit hosting agreement before release. Alternative models with clearly compatible deployment licenses may be simpler than negotiating uncertainty, but their voice-sample rights still require independent clearance.
Evidence, availability and commercial-rights boundary
Research cutoff: 25 October 2026. Provider documentation is evidence of its published interface or policy, not independent product testing. No performance score, hands-on trial, customer consent, territory-wide entitlement or legal clearance is claimed. Confirm the exact version, selected region, voice rights, data flow, account limits and current service terms before deployment. Unknown model release dates, account prices and plan limits are deliberately left unasserted; monetary claims belong to the separately dated AI10 pricing ledger. Where documentation is undated or dynamic, the public page is not treated as an archival snapshot of every earlier tier.