Loading...
Product
Core Features
Top Extensions
Company
State-of-the-art Chinese & English TTS — directly from your keyboard, 100% free during the public beta.
Read any selected or clipboard text aloud, design a brand-new voice from a single sentence, or clone any voice from a 10-second sample — without leaving Raycast.
Xiaomi's MiMo platform launched MiMo-V2.5-TTS Series + V2.5-ASR as a single full-stack speech suite, and the entire TTS family is currently limited-time free on their billing page:
| Model | What it does | Beta price |
|---|---|---|
mimo-v2.5-tts | Preset-voice synthesis with full style control | Free |
mimo-v2.5-tts-voicedesign | Generate a new voice from one sentence of description | Free |
mimo-v2.5-tts-voiceclone | Clone any voice from a small mp3/wav sample | Free |
mimo-v2-tts | Legacy voices | Free |
(Source: platform.xiaomimimo.com/docs/zh-CN/price/pay-as-you-go, as of 2026-05-27)
If you've been priced out of OpenAI Speech, ElevenLabs, or Azure Neural TTS for long-form reading or scripted character work, MiMo is the cheapest path to top-tier TTS today.
On 27 May 2026, Xiaomi shipped a permanent price cut across the MiMo-V2.5 LLM line — up to 99% off, with Token Plan quotas multiplied by 5–8× at the same price, plus a full credit reset for existing subscribers. (Announcement) Even after the TTS free beta ends, the underlying inference stack and billing engine just got a generational cost reduction. SGLang HiCache with SWA dropped KV-cache traffic to ~1/7 and cacheable tokens up ~5× — so unit cost is structurally lower, not just promotional.
MiMo-V2.5-TTS is positioned by Xiaomi as a model that "doesn't just read — it performs." All three creation modes share the same instruction-following surface:
(轻笑), (深呼吸), (粤语), (唱歌), (suppressed anger) mid-text for surgical control. Mix Chinese and English tags. Singing mode included.Most "Chinese TTS" services are awful at English; most "English TTS" services give you mechanical Mandarin. MiMo trains both at native quality:
Type a 1–4 sentence description and get a brand new voice synthesized on the spot:
"A weathered Northern Chinese grandfather, slow and steady pace, slightly raspy and time-worn, like he's telling old stories."
Optional optimize_text_preview flag lets the model auto-rewrite your sample text to match the persona — you can submit with an empty text body and MiMo writes the script for you. Other commercial TTS services charge you per generated voice ID; here it's one HTTP call.
Drop in any mp3 or wav (≤10 MB after base64). MiMo replicates the timbre and the cloned voice keeps the full control surface: director prompts, inline tags, dialect switching, singing mode. No upload step, no separate voice-management dashboard — each clone is a one-shot inline call.
afplay PID.This extension is published on the Raycast Store as MiMo TTS. Search for it in Raycast → Store, then:
tp-...) from https://platform.xiaomimimo.com/. (Pay-as-you-go sk- keys go to a different endpoint and are not used by this extension.)| Command | Mode | What it does |
|---|---|---|
| Quick Read | no-view | Read selected or clipboard text with your default voice. Trigger again to stop. |
| Read with Voice | view | Browse voices and read selection / clipboard with the chosen one. |
| Set Quick Read Voice | view | Pick and preview the voice that Quick Read uses. |
| TTS Studio | view | Long-form composer with voice, speed, opening style, emotion / rhythm / vocal-texture / expression tags, performance presets, and a free-form director prompt. |
| Design Voice | view | Generate a brand-new voice from a one-sentence description (MiMo-V2.5-TTS-VoiceDesign). |
| Clone Voice | view | Replicate any voice from an mp3/wav file (MiMo-V2.5-TTS-VoiceClone). |
| Setup Voice Defaults | view | Persist a per-session override for model / voice / rate / style prompt / Token Plan base URL. |
| Stop Reading | no-view | Stop the current playback immediately. |
| Speed up Reading / Slow Down Reading | no-view | Adjust playback speed by ±0.25× for the next chunk; persists globally. |
| Reading Status | menu-bar | Now-playing status with playback / speed controls. |
The TTS Studio command exposes everything MiMo-V2.5-TTS supports:
唱歌 overrides others to enter singing mode).| Model ID | Used by | What it does |
|---|---|---|
mimo-v2.5-tts | Quick Read · Read with Voice · Set Quick Read Voice · TTS Studio | Preset voices with style controls. |
mimo-v2.5-tts-voicedesign | Design Voice | Generate a voice from a 1–4 sentence description. Optional optimize_text_preview. |
mimo-v2.5-tts-voiceclone | Clone Voice | Replicate a voice from a base64-encoded mp3/wav (≤10 MB). |
mimo-v2-tts | optional, via Setup Voice Defaults | Legacy V2 voices. |
Quick Read uses one keystroke to start and stop. When something is already playing, running Quick Read again terminates the afplay process, clears the now-playing state, and shows a stop HUD. The menu-bar status, dedicated Stop Reading command, and cmd+. from any view command all trigger the same stop path.
Cross-extension PID isolation: this extension uses raycast-mimo-tts.pid / .stop in tmpdir so it doesn't fight my multi-provider AI Voice Studio extension over the same afplay process.
The extension reads selected text and the clipboard only when you trigger a command. Synthesized audio plays via the system afplay binary. No data is persisted beyond per-session Raycast LocalStorage (voice override, speech-rate override, now-playing state). The API key never leaves Raycast Preferences. Voice-clone samples are sent inline to MiMo's endpoint for that single request — no upload service is used and nothing is cached server-side per their docs.
Originally part of AI Voice Studio, a multi-provider Raycast TTS extension covering Qwen-TTS, MiniMax, MiMo, and OpenAI. This standalone version extracts the MiMo provider so users who only want MiMo TTS get a focused, smaller surface — no Qwen / MiniMax / OpenAI code paths.
Open source on GitHub: https://github.com/xwzhangSZU/Raycast-Mimo-TTS.
Decrease reading speed by 0.25× for the next playback.
Menu-bar status of the current MiMo TTS reading with playback and speed controls.
Choose and preview the MiMo voice used by Quick Read.
Generate a custom MiMo voice from a one-sentence description (MiMo-V2.5-TTS-VoiceDesign). Supports auto-optimized sample text.
Stop the current MiMo TTS playback.
Read selected text, or clipboard text if no selection is available. Trigger again to stop.
Generate and play MiMo speech from typed, selected, or pasted text, with voice, emotion, and style controls.
Browse MiMo voices and read selected or clipboard text aloud.
Replicate a voice from an mp3 or wav sample (≤10 MB) and read your text aloud (MiMo-V2.5-TTS-VoiceClone).
Configure the default MiMo model, voice, speech rate, style prompt, and token-plan base URL.
Increase reading speed by 0.25× for the next playback.