# ComfyUI-Omnivoice

A ComfyUI custom node for [OmniVoice](https://github.com/k2-fsa/OmniVoice) — a massive multilingual zero-shot TTS model supporting 600+ languages.

## Features

- **Voice Cloning** — clone any voice from a short reference audio clip
- **Voice Design** — describe a voice with text (e.g. "female, low pitch, british accent")
- **Auto Voice** — let the model pick a voice automatically
- **Voice Presets** — built-in curated reference voices, ready to use without any audio file
- **Voice Mixing** — blend two or three reference voices for a hybrid speaker
- **EPUB Loader** — load chapters from an ebook directly into the pipeline
- **Web Novel Loader** — load NovelFull chapters directly from a novel URL, with local caching
- **Audiobook-ready** — handles arbitrarily long text with near-constant VRAM via built-in chunking
- **Multilingual** — 600+ languages

## Installation

1. Clone into your ComfyUI custom nodes directory:
   ```bash
   cd ComfyUI/custom_nodes
   git clone https://github.com/ethanfel/ComfyUI-Omnivoice.git
   ```

2. Install via **ComfyUI Manager** (recommended) — it runs `install.py` and `requirements.txt` automatically.

   Or manually:
   ```bash
   pip install omnivoice --no-deps
   pip install -r requirements.txt
   ```
   > **Why `--no-deps` for omnivoice?** It pins `torch==2.8.*` from a CUDA 12.8 index. Installing it normally would overwrite ComfyUI's torch build. `install.py` handles this automatically; `requirements.txt` covers the remaining deps safely.

3. Restart ComfyUI. The nodes will appear under the **OmniVoice** category.

## Nodes

### OmniVoice Model Loader

Loads the OmniVoice model. Downloads automatically from HuggingFace on first run and caches to `ComfyUI/models/omnivoice/`.

| Input | Type | Description |
|-------|------|-------------|
| `device` | dropdown | `cuda:0`, `cuda:1`, or `cpu` |
| `dtype` | dropdown | `float16`, `bfloat16`, or `float32` |

**Output:** `OMNIVOICE_MODEL`

---

### OmniVoice Generate

Generates speech from text using a loaded model.

| Input | Type | Description |
|-------|------|-------------|
| `model` | OMNIVOICE_MODEL | From OmniVoice Model Loader |
| `text` | string | Text to synthesize (full pages supported) |
| `mode` | dropdown | `voice_cloning`, `voice_design`, or `auto_voice` |
| `ref_audio` | AUDIO | Reference audio for voice cloning (optional) |
| `ref_text` | string | Transcription of ref audio — connect a Whisper node for best results (optional) |
| `instruct` | string | Voice description for voice design mode (optional) |
| `speed` | float | Speed multiplier — default 1.0 |
| `num_step` | int | Diffusion steps — default 32 (use 16 for faster generation) |
| `seed` | int | Diffusion seed — set the same value across all Generate nodes in an audiobook pipeline to keep the voice consistent. 0 = random |

**Output:** `AUDIO` at 24kHz — connects directly to ComfyUI's Save Audio node.

---

### OmniVoice Voice Preset

Pre-fetched reference voices. Audio is downloaded once and cached to `ComfyUI/models/omnivoice/presets/`.

| Input | Type | Description |
|-------|------|-------------|
| `preset` | dropdown | Choose from built-in voices |

**Outputs:** `ref_audio` (AUDIO), `ref_text` (STRING) — wire directly into OmniVoice Generate.

Available presets:

| Name | Gender | Style |
|------|--------|-------|
| Shadowheart | Female | Expressive |
| American actress | Female | Theatrical |
| Podcast host | Female | Casual |
| Nature | Male | Warm |
| Old Hollywood | Male | Classic |
| Rick Sanchez | Male | Casual |
| Stewie Griffin | Male | Precise |
| Harvey Keitel | Male | Intense |
| Conan O'Brien | Male | Comedy |

---

### OmniVoice Mix Voices

Concatenates two or three reference audio clips to create a blended speaker. The model extracts a speaker embedding from the combined clip, producing a hybrid voice.

| Input | Type | Description |
|-------|------|-------------|
| `audio_1` | AUDIO | First reference voice (required) |
| `audio_2` | AUDIO | Second reference voice (required) |
| `weight_1` | float | Duration weight for audio_1 (0.0–1.0) |
| `weight_2` | float | Duration weight for audio_2 (0.0–1.0) |
| `audio_3` | AUDIO | Third reference voice (optional) |
| `weight_3` | float | Duration weight for audio_3 (optional) |
| `text_1/2/3` | string | Transcripts for each clip — merged into ref_text output |

**Outputs:** `ref_audio` (AUDIO), `ref_text` (STRING) — wire directly into OmniVoice Generate.

Weight controls how much of each clip's duration ends up in the mix. Equal weights (1.0 / 1.0) is a good starting point.

---

### OmniVoice EPUB Loader

Loads an EPUB file and outputs selected chapters as plain text, ready to pipe into OmniVoice Generate.

| Input | Type | Description |
|-------|------|-------------|
| `epub_path` | string | Absolute path to the `.epub` file |
| `chapter_start` | int | First chapter to include (1-indexed) |
| `chapter_end` | int | Last chapter to include (inclusive, auto-clamped) |

**Outputs:** `text` (STRING) — selected chapters joined by `---`, `chapter_list` (STRING) — numbered list of all chapter titles for reference.

### OmniVoice Web Novel Loader

Loads chapters directly from a NovelFull novel URL, such as
`https://novelfull.net/release-that-witch.html`. No EPUB conversion is needed.

| Input | Description |
|-------|-------------|
| `novel_url` | NovelFull's main novel page, rather than an individual chapter page |
| `chapter_start` | First chapter in reading order, starting at 1 |
| `chapter_end` | Last chapter, inclusive; connect the same counter to both inputs for one chapter per audio file |
| `refresh_cache` | Re-download the index and selected chapters; leave off for normal batch generation |

**Outputs:** `text`, `chapter_title`, `chapter_list` — the same outputs as the EPUB loader.
The body excludes the leading chapter heading and translator/editor credits, and filters
injected `ReadNovelFull.com` lines (including `freёReadNovelFull.com`). Connect
`chapter_title` to your existing title-and-body join to speak the title once.

The first run reads all pages of the chapter index. Only the selected chapter bodies
are downloaded. The index and chapter text are cached under
`ComfyUI/models/omnivoice/web_novels/`, so previously loaded chapters can be reused
offline. A counter past the final chapter raises an error instead of repeating the
last chapter. Currently, the supported website is `novelfull.net`.

The included `workflows/novelfull_release_that_witch.json` adapts the existing
chapter-by-chapter audiobook workflow:

- An incrementing integer feeds both chapter inputs, starting at 1.
- The title is joined with the body once, preserving the existing text filters.
- The voice reference, generation settings, and FLAC output are preserved.
- The filename prefix follows `Release That Witch/<chapter index> <chapter title>`.

Restart ComfyUI after updating the extension, load the workflow, and set the integer
counter to your starting chapter. Queue one run per desired chapter. The reference
audio must still be available at the path selected by the Load Audio node.

## Default Workflow

A ready-to-use workflow is included at `workflows/voice_cloning.json`:

```
[OmniVoice Model Loader] ──────────────────────────────────┐
                                                            ▼
[OmniVoice Voice Preset] ──► ref_audio ──► [OmniVoice Generate] ──► [Save Audio]
                         └──► ref_text ──►
```

Load it via ComfyUI → Load Workflow.

## Audiobook Pipeline

For multi-chapter audiobooks, use the same seed across all Generate nodes to keep the voice consistent between paragraphs:

```
[OmniVoice Model Loader] ──────────────────────────────────────────┐
                                                                    ▼
[OmniVoice EPUB Loader] ──► chapter text ──► [OmniVoice Generate] ──► [Save Audio]
                                                      ▲
[OmniVoice Voice Preset] ──► ref_audio / ref_text ──►
                              seed = 42 (fixed)
```

## Credits

- [OmniVoice](https://github.com/k2-fsa/OmniVoice) by k2-fsa
- [OmniVoice paper](https://arxiv.org/abs/2604.00688)