Fish Audio: Text to Speech, Voice Cloning, and Speech to Text in ComfyUI

Fish Audio is a speech AI service covering text to speech, voice cloning, and speech to text. In ComfyUI it ships as four partner nodes: Text to Speech, Speech to Text, Voice Selector, and Instant Voice Clone. This page walks through the ready-made workflow templates and how to connect the nodes. For the full inline tag reference (emotions, tone, pauses, phonemes), see the Fish Audio prompt guide.

​

Use Fish Audio in ComfyUI

The Fish Audio partner nodes are available in the node menu under partner/audio/Fish Audio. Four nodes cover the workflow:

• Fish Audio Text to Speech: converts text to speech, outputs AUDIO
• Fish Audio Speech to Text: transcribes audio to text, outputs text, language_code, and segments_json
• Fish Audio Voice Selector: selects one of the preset voices, or connects any fish.audio voice model through custom (the ID in https://fish.audio/m/<id>/), outputs voice
• Fish Audio Instant Voice Clone: creates a private cloned voice from recordings, outputs voice

​

Choosing a model

The model parameter on the Text to Speech node offers two options:

• s2.1-pro (recommended): supports multiple speakers through an autogrow list of 0 to 5 voice inputs
• s1 (legacy): accepts a single optional voice input and uses the parenthesis emotion tag syntax described in the prompt guide

​

Connecting voices

The voice output of the Voice Selector or the Instant Voice Clone connects to the Text to Speech node. The input name depends on the model: s2.1-pro (recommended) exposes the growable voices input (up to 5 voice slots), while s1 has a single optional voice input. With one connected voice, the node passes it internally as reference_id. With two or more, the text must mark speakers with @Voice1, @Voice2, and so on, and every connected voice must appear in the text at least once or the node raises an error. See Multi-speaker dialogue in the prompt guide.

​

Voice cloning limits

The Instant Voice Clone node accepts 1 to 20 recordings, recommends 10 to 30 seconds per clip, and validates that the total duration stays under 270 seconds. The enhance_audio_quality option is enabled by default. For recording tips, see Voice cloning quality in the prompt guide.

​

Account and pricing

The nodes require signing in to a ComfyUI account and are billed per use. Reference prices from the nodes: text to speech is about $21.45 per 1M bytes of text, speech to text is$0.00858 per minute, and voice cloning is free.

​

Available workflows

Three ready-made templates cover the main Fish Audio flows: text to speech, voice clone to speech, and speech to text. Open them from the Template Library in ComfyUI, or run them directly on Comfy Cloud. The voice clone and speech to text templates require ComfyUI 0.34.0 or newer.

​

Fish Audio: Text to Speech

Generate speech from your script with a preset voice. Write your text with the inline tags from the prompt guide, then pick another preset in the Voice Selector or connect a cloned voice. Input materials Upload this file to the matching LoadAudio node:

​

Fish Audio: Voice Clone to Speech

Clone a voice from a short recording, then use it to speak any script. Record 2 to 3 clips of 15 to 20 seconds, upload them through the LoadAudio node (or record directly in the RecordAudio node), and connect the Instant Voice Clone output to the Text to Speech node. For recording tips, see Voice cloning quality in the prompt guide. Input materials Upload this file to the matching LoadAudio node:

​

Fish Audio: Speech to Text

Transcribe an audio clip to text. Language is detected automatically across 80+ languages, and enabling precise_timestamps returns per-segment start and end times as JSON, which is handy for subtitles. Input materials Upload this file to the matching LoadAudio node:

​

Get started

1. Update ComfyUI to the latest version
2. Open a workflow template above, or add the Fish Audio nodes from the node menu under partner/audio/Fish Audio
3. Sign in to your ComfyUI account when prompted
4. Write your script with the inline tags from the prompt guide
5. Run the workflow and preview or save the audio