Kling TTS Advanced

The plain Kling TTS node is "say this, in this voice." This one is the director's cut: same speech synthesis, but with a speed dial and a language setting, so the voice line can actually match your scene's pacing and language. You reach for it when the default delivery is too fast or too flat for a dramatic line, when you're working in a non-English language, or when you want to lock a consistent voice character across a project.

Four inputs, and three of them are what you'll fiddle with:

• text - multiline this time (the basic TTS is single-line), so longer voice lines don't fight the input box.
• voice_id - defaults to girlfriend_4_speech02 here, not the plain TTS's female_1. Use the pack's Voice Selector to browse presets or drop in a cloned voice's ID.
• voice_speed - 0.5 to 2.0, where 1.0 is normal. This is the reason to pick this node over the plain one: drop to 0.8 for a slower, weightier read, or push 1.3 for a frantic line.
• voice_language - a language code (default en; zh for Chinese, etc.). Set it to match your text; a mismatched code is how you get a voice that pronounces English with the wrong mouth.

The mechanism is identical to the plain TTS - one cloud call, submit-and-poll, speech comes back. Outputs are the same quartet: audio (AUDIO), audio_file, url, task_id. The audio output is the one you'll care about most; wire it into a preview node or feed it into the Lip Sync node for a fully-voiced character clip.

Installing it

Part of ComfyUI-API-Toolkit. Manager: search "API Toolkit". Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-API-Toolkit
pip install -r requirements.txt

Restart ComfyUI. As with all the pack's audio nodes, install the full requirements (soundfile, scipy included) so the AUDIO output actually decodes.

Gotchas

Voice pacing in AI TTS is a double-edged sword: voice_speed changes the tempo, but Kling won't re-act lines emotionally - a flat read at 0.7 is still a flat read, just slower. If the delivery itself is wrong, rephrase the text or pick a different voice rather than speed-sliding your way out. Also remember every queue is a billed TTS call, and this node makes it easy to burn a few credits auditioning speed values - settle on one and reuse it. And if you're pairing this with Lip Sync's text2video mode, note that Lip Sync has its own voice_speed knob; tune it there for the final render, not here, so you don't double-apply tempo.