Observed facts: Vercel says Drama 3 from Fish Audio is available on AI Gateway in preview, and gives fish-audio/drama-3 as the model name. The release describes natural-language directions for emotion and pacing, pause and delivery tags, directions in a language different from the spoken text, and separate voices for multi-speaker dialogue. It also advertises 80% off for 30 days through November 12.
Mechanism: The mechanism is to put performance direction in the text/script sent to a text-to-speech model, then invoke that model through the Gateway’s unified interface. This gives an application a way to request differentiated delivery and character voices without treating speech output as an undirected text-to-audio conversion. The announcement does not describe how these controls are implemented or expose comparative evaluations.
Why now: The availability announcement makes this a current integration option for teams already using AI Gateway, while the temporary discount creates a near-term reason to test cost and behavior. The release labels the model preview, so the change is access to an experiment, not evidence of general availability or production reliability.
Implications: For teams building narrated content or character dialogue, script-level direction may make generated speech more composable in an application workflow. The practical value will depend on whether the controls behave consistently across voices, languages, and longer scripts; those properties are not established here. A second implication is that gateway inclusion simplifies discovery and invocation, but does not itself validate the underlying model.
Evidence boundary: The evidence is Vercel’s changelog, which reports availability, supported controls, model identifier, preview status, and a time-limited discount. It provides no independent quality measurements, latency or reliability data, supported-language list, or production-readiness guarantee.
Watch: Test whether emotion, pacing, and speaker distinctions remain reliable across representative scripts and languages before making the model a production dependency.