ComfyUI: TTS WebUI (OpenAI-style) nodes

This package provides ComfyUI custom nodes that call a TTS WebUI OpenAI-style HTTP endpoint (POST /v1/audio/speech) and return in-memory AUDIO usable in ComfyUI workflows.

What it does

• Sends text + options to a TTS WebUI endpoint and requests WAV output.
• Converts the returned WAV bytes into a ComfyUI AUDIO object (torch.Tensor waveform + sample rate).
• Optionally saves the WAV to the ComfyUI output/audio folder and returns metadata JSON.

Nodes included

• TTS WebUI via API — class TTSWebUI_OpenAI_TTS (generic, "jack of all trades")
• TTS Kokoro — class TTS_KokoroNode
• TTS Chatterbox — class TTS_ChatterboxNode
• TTS StyleTTS2 — class TTS_StyleTTS2Node
• TTS Kitten TTS — class TTS_KittenTTSNode
• TTS F5-TTS — class TTS_F5TTSNode
• TTS Preset — class TTS_GlobalPresetNode

All nodes share a minimal helper (NodeTTSBase) for HTTP and response handling. Each node builds its own JSON payload explicitly to avoid hidden cross-file magic. Common required inputs text and api_base are provided via a small required_inputs() helper on the base class to reduce duplication.

Common required inputs

• text (STRING, multiline) — text to synthesize
• api_base (STRING) — base URL of your TTS WebUI (default: http://127.0.0.1:7778)

Common optional inputs

• api_key (STRING, password) — API key for the TTS server (if required)
• timeout_sec (INT) — request timeout in seconds
• channels_first (BOOLEAN) — AUDIO layout: (C, N) if True else (N, C)
• also_save_wav (BOOLEAN) — if True, saves a WAV under output/audio and returns the path
• save_prefix (STRING) — filename prefix for saved WAVs
• return_metadata (BOOLEAN) — if True, returns a metadata JSON string

Model-specific nodes add their own inputs (e.g., voice, speed, alpha/beta, etc.) and build the corresponding payload under the hood.

Advanced: raw payload override (generic node)

TTSWebUI_OpenAI_TTS has an advanced_request_json input to fully override the default payload. Example body sent to /v1/audio/speech:

{
  "model": "kokoro",
  "input": "Hello from ComfyUI",
  "voice": "random",
  "speed": 1.0,
  "response_format": "wav",
  "params": {"temperature": 0.7}
}

Installation / Requirements

1. Place this folder inside your ComfyUI custom_nodes directory.
2. Ensure the environment has these Python packages installed: requests, numpy, torch.
   • If your ComfyUI environment already has torch installed, no extra action is required.

API key detection

If your TTS server requires an API key you can either:

• pass it into the api_key input, or
• set the environment variable TTS_WEBUI_OPENAI_API_KEY to your API key (preferred).

For backward compatibility the module will also check SAI_API_KEY. As a local-file fallback the helper also reads tts_api_key.txt or sai_platform_key.txt placed next to the custom node. When a key is found, requests include the header Authorization: Bearer <key>.

Layout

• nodes/ — one node per file (generic + per-model)
• nodes/_base.py — NodeTTSBase providing HTTP and finalize helpers and required_inputs()
• utils/ — small helpers for audio I/O, paths, filenames
• __init__.py — package exports for ComfyUI

Notes

This repository originally included a Stability API template. The TTS nodes are independent of that code; any legacy modules (tts_base.py, tts_models.py, nodes/_shared.py) are now deprecated and will raise a clear ImportError if imported.

---

License: see LICENSE