音频｜豆包语音合成 2.0

If your workflow ends in spoken Chinese - product narration, short-video voiceovers, character dialogue - this is the node that stops you from hand-typing ByteDance's opaque voice_type IDs. TikpanDoubaoTTS20Node wraps Doubao's Seed-TTS 2.0 through the Tikpan relay, hides the fixed host, and hands you a 25-voice dropdown with human-readable labels like 小何 2.0 / 女 / 清新自然. The author's default text literally says it: "音色已经整理成下拉框，普通用户不用再手动查 voice_type" - the voices are the feature.

How it works

You type text, pick a voice and a few prosody sliders, and the node POSTs to Tikpan's bound /api/v3/tts/unidirectional/sse path (Doubao's V3 unidirectional SSE). The returned audio is saved locally and emitted three ways: 📁_音频路径 (file path), 🔗_音频链接 (hosted URL), and 🎧_音频流 (an AUDIO output you can wire straight into a video-composition node). Plus 🎙️_voice_type, 💰_计费字符数 (billed character count - billing is per character), and 📋_状态日志. It's an is_output_node (output-marked) node, but the 🎧_音频流 wire is what feeds downstream muxing.

The inputs that actually matter

• API_密钥 - sk- key from tikpan.com.
• 合成文本 - the text to speak. Per-character billing, so keep it tight.
• 音色 - 25 preset voices, each with gender/style annotated. Default zh_female_vv_uranus_bigtts (通用 VV 2.0 女声, multilingual + dialect support).
• 语速 / 音量 / 音调 - speed 0.5–2×, volume 0.1–3×, pitch 0.5–2×. The volume tooltip warns 过高可能爆音 (too loud clips).
• 情感 - optional emotion tag: happy / sad / angry / fearful / surprised / neutral.
• 音频格式 and 采样率 - mp3 (default) / wav / pcm, and 24k default sampling up to 48k.

Optional: 自定义voice_type (for voice cloning ICL or brand-new official voices), 资源ID (seed-tts-2.0 default; the ICL variants are the cloning path), 复用本地缓存 (default on - same text + params hits local cache and avoids a double charge, worth leaving on), and 跳过错误 (returns a silent clip on failure so batch flows survive).

Installing it

Same pack, same steps:

cd ComfyUI/custom_nodes
git clone https://github.com/htrert/ComfyUI-Tikpan-Pro

restart (or Manager → "Tikpan"). No models to download - this is pure relay + file plumbing.

Where people get burned

• Billing is per character. The 💰_计费字符数 output exists so you can watch spend. Long scripts are the real cost driver.
• Cert verification defaults ON here (true), unlike the new.ip233.com nodes - which is correct behavior, but if you're behind a MITM proxy you'll get cert errors and have to toggle it off. The tooltip says as much.
• The cache is a money-saver, not a bug. If a repeat run seems to return instantly with no charge, that's 复用本地缓存 working.
• Voice cloning is a hidden feature. ICL resource IDs + a custom voice_type is the path to voice replication - the dropdown won't tell you this, but the schema will.

The one-liner: if you need clean, natural Chinese TTS in a graph without memorizing ByteDance's ID scheme, this is the node. It's the "voices in a dropdown" version of Doubao TTS, and for Chinese-first content that's a real quality-of-life win.