ElevenLabs: Eleven Flash v2

Eleven Flash v2 is a text-to-speech model from ElevenLabs, listed on api14 under the speech category. It is an English-only model built for ultra-low-latency synthesis, which places it in the real-time and conversational part of the ElevenLabs lineup rather than among the company's higher-fidelity production voices.

The model takes text as its only input and returns speech as its only output. It does not accept image, audio or other non-text inputs, and it does not generate text. The model card lists no context length in tokens, so there is no token window to plan around; the practical input constraint is the 30,000-character request limit per call. Twenty-one named voices are available: george, sarah, adam, alice, bella, bill, brian, callum, charlie, chris, daniel, eric, harry, jessica, laura, liam, lily, matilda, river, roger and will. Voice choice is the main lever you have over the output, since no other generation parameters are exposed.

Those characteristics point to a fairly specific set of uses. Because latency is the priority and the model is English-only, it fits voice agents that answer and respond in a live conversation, spoken replies in chat or assistant interfaces, and read-aloud features where a pause of a second or more would be noticeable. Short prompts, confirmations, menu narration and turn-by-turn dialogue are natural fits, and the 30,000-character limit is generous for those cases. Longer narration, audiobooks and content where delivery quality matters more than response time are better served by other ElevenLabs models, and any workflow that needs a language other than English will need a different model entirely.

You call the model through api14's OpenAI-compatible API using the model id "elevenlabs/eleven-flash-v2". The request body carries your text and the voice you want. The model card lists no supported parameters beyond the voice list, so there is nothing else to tune: no temperature, no speed or style controls, and no sampling options. If your code already speaks the OpenAI chat or audio interface, pointing the model field at this id and supplying the text is the whole integration.

The main things to watch are language and scope. English-only means non-English text is out of scope, and the absence of tunable parameters means you cannot adjust delivery from the request; pick the right voice up front instead. If you need multilingual output, longer-form narration, or finer control over how the speech sounds, choose a different model.