s1.
What is changing
S1 is deprecated and will be retired on December 31, 2026. After that date, requests that specifys1 are served by s2.1-pro.
We recommend that you switch to s2.1-pro before then, so you can test your output on your own schedule.
What happens if you do nothing
After December 31, 2026, requests that specifymodel: s1 are automatically served by s2.1-pro and billed as s2.1-pro. Both models have the same price, $15.00 / M UTF-8 bytes, so your cost does not change. See Pricing & Rate Limits.
Your audio can still change, because S2.1-Pro treats some input and defaults differently than S1:
- S1-style
(happy)emotion tags are not interpreted as tags, and the API does not convert them. - Loudness normalization is applied to the output.
repetition_penaltyhas no effect.
Migrate to S2.1-Pro
Switch the model to s2.1-pro
In the API, you select the model with the model request header. Change s1 to s2.1-pro wherever you set the model, including SDK and integration settings.
If you use the JavaScript SDK (fish-audio on npm) without passing a model, your requests use s1, because that is the SDK’s default backend. Pass the model explicitly.
This example also converts the emotion tags, as described in the next step:
Convert emotion tags
S1 uses(parenthesis) tags, such as (happy). S2 uses [bracket] tags, such as [happy], and accepts free-form natural language, such as [whispers sweetly].
The API does not convert old tags. S1-style (happy) text sent to S2 is not interpreted as a tag, so replace each (tag) in your text with [tag]:
Generate multi-speaker dialogue in one request
With S1, multi-speaker audio meant generating each speaker’s segment separately and stitching the audio together. S2.1-Pro supports multi-speaker dialogue natively in one request: add<|speaker:N|> markers to text, and pass reference_id as an array with one voice ID per speaker. For details and an example, see reference_id in the Text to Speech API reference.
Check output loudness
normalize_loudness in the prosody object defaults to true. It had no effect on S1, but it is applied on S2, so your output loudness can change after you migrate.
If you need to turn it off, set prosody.normalize_loudness to the boolean false:
Check repetition_penalty
repetition_penalty applies to S1 but has no effect on S2.1-Pro. If you tuned it to reduce repeated sounds in S1 output, that setting no longer changes your audio after you switch. Listen to your output after you migrate.
Support
Need help? Check out these resources:- API Reference - Complete API documentation
- Create a Voice Clone - Create a voice clone model
- Generate Speech - Generate realistic speech
- Real-time Streaming - WebSocket for real-time streaming
- Discord Community - Get help from the community
- Support Email - Contact our support team