Migrating from Alania-1
Alania-2 is now the default model. Requests that name no model reach Alania-2. To stay on Alania-1, send model: "alania-v1".
What changed
| alania-v1 | alania-v2 | |
|---|---|---|
| Default | No (legacy) | Yes |
| Sample rate | 24 kHz | 48 kHz |
| Voices | alania | alania2-f1, alania2-oncel and more |
| Style, design, cloning | No | Yes |
| temperature | 0–2.0, default 0.30 | 0.6–2.0, default 1.0 |
| top_p · cfg | top_p | cfg |
| Price | $0.015 · ₺0,75 / 1,000 characters | The same |
Where your request goes
| You send | Now |
|---|---|
No model, no voice | Alania-2, alania2-f1, 48 kHz |
No model, voice: "alania" | Alania-2, alania2-f1, 48 kHz |
No model, voice: "alania-v1" | Alania-1, unchanged |
model: "alania-v1" | Alania-1, unchanged |
model: "tts-1" (OpenAI names) | Alania-2 |
Staying on Alania-1
On a WebSocket, open the connection with ?model=alania-v1, or add model: "alania-v1" to each speak message.
Moving to Alania-2
- Send
model: "alania-v2"and pick a voice:alania2-f1,alania2-oncelor another built-in voice. - Make your player ready for 48 kHz; read the rate from the response.
- If you send
temperature, retune it: Alania-2's default is 1.0.top_pno longer has an effect. - Optionally add style instructions.