Fish Audio S

Fish Audio's bet is narrow and clear: keep the model's emotional range high and the entry price at zero, then upsell volume. The thing that actually differentiates it is the tag grammar. Emotion tags and effect tags are first-class in the editor and in the API — [angry], [whispering], [breathy] sit alongside [laughing], [sobbing], [sighing], [panting] and [long pause] — which means directors and game teams can direct a line the way they'd annotate a script rather than re-recording until the model guesses right. Very few TTS products let you type [embarrassed] and hear the difference. The second differentiator is price architecture. The Free tier is genuinely usable for prototyping: 8,000 credits/month, roughly 600-625 credits per minute of generation, so about 7 minutes of output. Credits reset monthly and unused minutes do not roll over, which is the main gotcha on the free and low tiers. The jump from Free to Plus ($11/mo billed annually, $15 month-to-month) is where the real product opens up — 250,000 credits, up to 200 minutes, 15,000 characters per generation, private voice slots, priority generation on the newest models and Voice Design. Pro at $75/mo billed annually ($100 month-to-month) is the tier most working teams land on: 2,000,000 credits, up to 1,620 minutes, 3 team seats, unlimited voice slots and 5 professional (verified) voice slots, plus a 7-day money-back guarantee. Max at $749/mo billed annually ($999 month-to-month) scales to 25,000,000 credits, roughly 6,250 minutes, 10 seats and 15 professional voice slots. Where it gets thin is enterprise governance. Zero Data Retention, on-premise deployment and SOC 2 compliance are Enterprise-only and priced by contact. Custom SSO is listed as coming soon, not shipped. If procurement requires those controls before a pilot, you're having a sales conversation, not a signup. Commercial rights are also tier-gated: Free plan output is personal, non-commercial only; commercial use starts at Plus with verified voices you own. The 2,000,000+ voice library is a genuine asset — it means you can audition castings instead of building them — but it also means quality varies, and the ethical and legal status of user-uploaded voices is something you should think about before shipping a public product. The platform is browser-based; there is no offline or desktop mode documented. Supported languages are named for English, Japanese, Korean, Chinese, French, German, Arabic and Spanish within the 30+ set, and tag behavior across every language and voice is not guaranteed. Clone fidelity depends on clean input audio — 15 seconds of noisy phone recording won't give you the same result as a treated studio take. Where it fits: solo YouTubers and explainer channels, audiobook producers who need ACX/Audible-ready narration without a booth, indie game and animation teams directing per-line emotion, and developers prototyping conversational agents against a free real-time API. Where it doesn't: enterprises